Experimental: display keyframe interpolation mode in dopesheet.

Allow distinguishing 7 basic interpolation classes. Currently the
dopesheet allows changing interpolation and handle types, but does
not show the current type in any way.

.gitignore

@@ -21,7 +21,6 @@ Desktop.ini
 
 # commonly used paths in blender
 /blender.bin
-/user-config.py
 /BUILD_NOTES.txt
 
 # local patches
@@ -34,8 +33,3 @@ Desktop.ini
 /doc/python_api/sphinx-in/
 /doc/python_api/sphinx-out/
 /doc/python_api/rst/bmesh.ops.rst
-/doc/python_api/rst/in_menu.png
-/doc/python_api/rst/menu_id.png
-/doc/python_api/rst/op_prop.png
-/doc/python_api/rst/run_script.png
-/doc/python_api/rst/spacebar.png

.gitmodules

@@ -2,7 +2,7 @@
 	path = release/scripts/addons
 	url = ../blender-addons.git
 	ignore = all
-	branch = master
+	branch = blender-v2.79-release
 [submodule "release/scripts/addons_contrib"]
 	path = release/scripts/addons_contrib
 	url = ../blender-addons-contrib.git
@@ -12,9 +12,9 @@
 	path = release/datafiles/locale
 	url = ../blender-translations.git
 	ignore = all
-	branch = master
+	branch = blender-v2.79-release
 [submodule "source/tools"]
 	path = source/tools
 	url = ../blender-dev-tools.git
 	ignore = all
-	branch = master
+	branch = blender-v2.79-release

CMakeLists.txt

@@ -32,7 +32,7 @@ if(${CMAKE_SOURCE_DIR} STREQUAL ${CMAKE_BINARY_DIR})
 	if(NOT DEFINED WITH_IN_SOURCE_BUILD)
 		message(FATAL_ERROR
 			"CMake generation for blender is not allowed within the source directory!"
-			"\n Remove the CMakeCache.txt file and try again from another folder, e.g.:"
+			"\n Remove \"${CMAKE_SOURCE_DIR}/CMakeCache.txt\" and try again from another folder, e.g.:"
 			"\n "
 			"\n rm CMakeCache.txt"
 			"\n cd .."
@@ -58,7 +58,9 @@ list(APPEND CMAKE_MODULE_PATH "${CMAKE_SOURCE_DIR}/build_files/cmake/Modules")
 list(APPEND CMAKE_MODULE_PATH "${CMAKE_SOURCE_DIR}/build_files/cmake/platform")
 
 # avoid having empty buildtype
-set(CMAKE_BUILD_TYPE_INIT "Release")
+if(NOT DEFINED CMAKE_BUILD_TYPE_INIT)
+	set(CMAKE_BUILD_TYPE_INIT "Release")
+endif()
 
 # quiet output for Makefiles, 'make -s' helps too
 # set_property(GLOBAL PROPERTY RULE_MESSAGES OFF)
@@ -229,7 +231,7 @@ mark_as_advanced(BUILDINFO_OVERRIDE_TIME)
 
 option(WITH_IK_ITASC      "Enable ITASC IK solver (only disable for development & for incompatible C++ compilers)" ON)
 option(WITH_IK_SOLVER     "Enable Legacy IK solver (only disable for development)" ON)
-option(WITH_FFTW3         "Enable FFTW3 support (Used for smoke and audio effects)" ${_init_FFTW3})
+option(WITH_FFTW3         "Enable FFTW3 support (Used for smoke, ocean sim, and audio effects)" ${_init_FFTW3})
 option(WITH_BULLET        "Enable Bullet (Physics Engine)" ON)
 option(WITH_SYSTEM_BULLET "Use the systems bullet library (currently unsupported due to missing features in upstream!)" )
 mark_as_advanced(WITH_SYSTEM_BULLET)
@@ -1380,10 +1382,20 @@ if(CMAKE_COMPILER_IS_GNUCC)
 		ADD_CHECK_CXX_COMPILER_FLAG(CXX_WARNINGS CXX_WARN_MISSING_DECLARATIONS -Wmissing-declarations)
 	endif()
 
+	# Use 'ATTR_FALLTHROUGH' macro to suppress.
+	if(CMAKE_COMPILER_IS_GNUCC AND (NOT "${CMAKE_C_COMPILER_VERSION}" VERSION_LESS "7.0"))
+		ADD_CHECK_C_COMPILER_FLAG(C_WARNINGS C_WARN_IMPLICIT_FALLTHROUGH -Wimplicit-fallthrough=5)
+		ADD_CHECK_CXX_COMPILER_FLAG(CXX_WARNINGS CXX_WARN_IMPLICIT_FALLTHROUGH -Wimplicit-fallthrough=5)
+	endif()
+
 	# flags to undo strict flags
 	ADD_CHECK_C_COMPILER_FLAG(CC_REMOVE_STRICT_FLAGS C_WARN_NO_DEPRECATED_DECLARATIONS -Wno-deprecated-declarations)
 	ADD_CHECK_C_COMPILER_FLAG(CC_REMOVE_STRICT_FLAGS C_WARN_NO_UNUSED_PARAMETER        -Wno-unused-parameter)
 
+	if(CMAKE_COMPILER_IS_GNUCC AND (NOT "${CMAKE_C_COMPILER_VERSION}" VERSION_LESS "7.0"))
+		ADD_CHECK_C_COMPILER_FLAG(CC_REMOVE_STRICT_FLAGS C_WARN_NO_IMPLICIT_FALLTHROUGH    -Wno-implicit-fallthrough)
+	endif()
+
 	if(NOT APPLE)
 		ADD_CHECK_C_COMPILER_FLAG(CC_REMOVE_STRICT_FLAGS C_WARN_NO_ERROR_UNUSED_BUT_SET_VARIABLE -Wno-error=unused-but-set-variable)
 	endif()

GNUmakefile

@@ -104,7 +104,7 @@ endif
 CMAKE_CONFIG = cmake $(BUILD_CMAKE_ARGS) \
                      -H"$(BLENDER_DIR)" \
                      -B"$(BUILD_DIR)" \
-                     -DCMAKE_BUILD_TYPE:STRING=$(BUILD_TYPE)
+                     -DCMAKE_BUILD_TYPE_INIT:STRING=$(BUILD_TYPE)
 
 
 # -----------------------------------------------------------------------------

build_files/build_environment/install_deps.sh

@@ -26,7 +26,7 @@ ARGS=$( \
 getopt \
 -o s:i:t:h \
 --long source:,install:,tmp:,info:,threads:,help,show-deps,no-sudo,no-build,no-confirm,use-cxx11,\
-with-all,with-opencollada,\
+with-all,with-opencollada,with-jack,\
 ver-ocio:,ver-oiio:,ver-llvm:,ver-osl:,ver-osd:,ver-openvdb:,\
 force-all,force-python,force-numpy,force-boost,\
 force-ocio,force-openexr,force-oiio,force-llvm,force-osl,force-osd,force-openvdb,\
@@ -118,6 +118,9 @@ ARGUMENTS_INFO="\"COMMAND LINE ARGUMENTS:
     --with-opencollada
         Build and install the OpenCOLLADA libraries.
 
+    --with-jack
+        Install the jack libraries.
+
     --ver-ocio=<ver>
         Force version of OCIO library.
 
@@ -289,7 +292,7 @@ NO_BUILD=false
 NO_CONFIRM=false
 USE_CXX11=false
 
-PYTHON_VERSION="3.5.2"
+PYTHON_VERSION="3.5.3"
 PYTHON_VERSION_MIN="3.5"
 PYTHON_FORCE_BUILD=false
 PYTHON_FORCE_REBUILD=false
@@ -322,8 +325,8 @@ OPENEXR_FORCE_REBUILD=false
 OPENEXR_SKIP=false
 _with_built_openexr=false
 
-OIIO_VERSION="1.7.13"
-OIIO_VERSION_MIN="1.7.13"
+OIIO_VERSION="1.7.15"
+OIIO_VERSION_MIN="1.7.15"
 OIIO_VERSION_MAX="1.9.0"  # UNKNOWN currently # Not supported by current OSL...
 OIIO_FORCE_BUILD=false
 OIIO_FORCE_REBUILD=false
@@ -366,8 +369,7 @@ ALEMBIC_FORCE_BUILD=false
 ALEMBIC_FORCE_REBUILD=false
 ALEMBIC_SKIP=false
 
-# Version??
-OPENCOLLADA_VERSION="1.3"
+OPENCOLLADA_VERSION="1.6.51"
 OPENCOLLADA_FORCE_BUILD=false
 OPENCOLLADA_FORCE_REBUILD=false
 OPENCOLLADA_SKIP=false
@@ -508,6 +510,9 @@ while true; do
     --with-opencollada)
       WITH_OPENCOLLADA=true; shift; continue
     ;;
+    --with-jack)
+      WITH_JACK=true; shift; continue;
+    ;;
     --ver-ocio)
       OCIO_VERSION="$2"
       OCIO_VERSION_MIN=$OCIO_VERSION
@@ -711,6 +716,9 @@ done
 if [ "$WITH_ALL" = true -a "$OPENCOLLADA_SKIP" = false ]; then
   WITH_OPENCOLLADA=true
 fi
+if [ "$WITH_ALL" = true ]; then
+  WITH_JACK=true
+fi
 
 
 WARNING "****WARNING****"
@@ -737,7 +745,10 @@ _boost_version_nodots=`echo "$BOOST_VERSION" | sed -r 's/\./_/g'`
 BOOST_SOURCE=( "http://sourceforge.net/projects/boost/files/boost/$BOOST_VERSION/boost_$_boost_version_nodots.tar.bz2/download" )
 BOOST_BUILD_MODULES="--with-system --with-filesystem --with-thread --with-regex --with-locale --with-date_time --with-wave --with-iostreams --with-python --with-program_options"
 
+OCIO_USE_REPO=true
 OCIO_SOURCE=( "https://github.com/imageworks/OpenColorIO/tarball/v$OCIO_VERSION" )
+OCIO_SOURCE_REPO=( "https://github.com/imageworks/OpenColorIO.git" )
+OCIO_SOURCE_REPO_UID="6de971097c7f552300f669ed69ca0b6cf5a70843"
 
 OPENEXR_USE_REPO=false
 OPENEXR_SOURCE=( "http://download.savannah.nongnu.org/releases/openexr/openexr-$OPENEXR_VERSION.tar.gz" )
@@ -786,7 +797,7 @@ ALEMBIC_SOURCE=( "https://github.com/alembic/alembic/archive/${ALEMBIC_VERSION}.
 # ALEMBIC_SOURCE_REPO_BRANCH="master"
 
 OPENCOLLADA_SOURCE=( "https://github.com/KhronosGroup/OpenCOLLADA.git" )
-OPENCOLLADA_REPO_UID="3335ac164e68b2512a40914b14c74db260e6ff7d"
+OPENCOLLADA_REPO_UID="0c2cdc17c22cf42050e4d42154bed2176363549c"
 OPENCOLLADA_REPO_BRANCH="master"
 
 FFMPEG_SOURCE=( "http://ffmpeg.org/releases/ffmpeg-$FFMPEG_VERSION.tar.bz2" )
@@ -1268,7 +1279,7 @@ compile_OCIO() {
   fi
 
   # To be changed each time we make edits that would modify the compiled result!
-  ocio_magic=1
+  ocio_magic=2
   _init_ocio
 
   # Clean install if needed!
@@ -1285,14 +1296,27 @@ compile_OCIO() {
     if [ ! -d $_src ]; then
       INFO "Downloading OpenColorIO-$OCIO_VERSION"
       mkdir -p $SRC
-      download OCIO_SOURCE[@] $_src.tar.gz
 
-      INFO "Unpacking OpenColorIO-$OCIO_VERSION"
-      tar -C $SRC --transform "s,(.*/?)imageworks-OpenColorIO[^/]*(.*),\1OpenColorIO-$OCIO_VERSION\2,x" \
-          -xf $_src.tar.gz
+      if [ "$OCIO_USE_REPO" = true ]; then
+        git clone ${OCIO_SOURCE_REPO[0]} $_src
+      else
+        download OCIO_SOURCE[@] $_src.tar.gz
+        INFO "Unpacking OpenColorIO-$OCIO_VERSION"
+        tar -C $SRC --transform "s,(.*/?)imageworks-OpenColorIO[^/]*(.*),\1OpenColorIO-$OCIO_VERSION\2,x" \
+            -xf $_src.tar.gz
+      fi
+
     fi
 
     cd $_src
+
+    if [ "$OCIO_USE_REPO" = true ]; then
+      # XXX For now, always update from latest repo...
+      git pull origin master
+      git checkout $OCIO_SOURCE_REPO_UID
+      git reset --hard
+    fi
+
     # Always refresh the whole build!
     if [ -d build ]; then
       rm -rf build
@@ -1498,7 +1522,6 @@ compile_OPENEXR() {
     if [ "$OPENEXR_USE_REPO" = true ]; then
       # XXX For now, always update from latest repo...
       git pull origin master
-      # Stick to same rev as windows' libs...
       git checkout $OPENEXR_SOURCE_REPO_UID
       git reset --hard
       oiio_src_path="../OpenEXR"
@@ -1576,7 +1599,7 @@ compile_OIIO() {
   fi
 
   # To be changed each time we make edits that would modify the compiled result!
-  oiio_magic=16
+  oiio_magic=17
   _init_oiio
 
   # Clean install if needed!
@@ -1640,6 +1663,9 @@ compile_OIIO() {
       INFO "ILMBASE_HOME=$INST/openexr"
     fi
 
+    # ptex is only needed when nicholas bishop is ready
+    cmake_d="$cmake_d -D USE_PTEX=OFF"
+
     # Optional tests and cmd tools
     cmake_d="$cmake_d -D USE_QT=OFF"
     cmake_d="$cmake_d -D USE_PYTHON=OFF"
@@ -2638,7 +2664,7 @@ install_DEB() {
     PRINT ""
   fi
 
-  if [ "$WITH_ALL" = true ]; then
+  if [ "$WITH_JACK" = true ]; then
     _packages="$_packages libspnav-dev"
     # Only install jack if jack2 is not already installed!
     JACK="libjack-dev"
@@ -3175,7 +3201,7 @@ install_RPM() {
   if [ "$RPM" = "FEDORA" -o "$RPM" = "RHEL" ]; then
     _packages="$_packages freetype-devel tbb-devel"
 
-    if [ "$WITH_ALL" = true ]; then
+    if [ "$WITH_JACK" = true ]; then
       _packages="$_packages jack-audio-connection-kit-devel"
     fi
 
@@ -3649,7 +3675,11 @@ install_ARCH() {
   THEORA_USE=true
 
   if [ "$WITH_ALL" = true ]; then
-    _packages="$_packages jack libspnav"
+    _packages="$_packages libspnav"
+  fi
+
+  if [ "$WITH_JACK" = true ]; then
+    _packages="$_packages jack"
   fi
 
   PRINT ""
@@ -4311,6 +4341,14 @@ print_info() {
     _buildargs="$_buildargs $_1"
   fi
 
+  if [ "$WITH_JACK" = true ]; then
+    _1="-D WITH_JACK=ON"
+    _2="-D WITH_JACK_DYNLOAD=ON"
+    PRINT "  $_1"
+    PRINT "  $_2"
+    _buildargs="$_buildargs $_1 $_2"
+  fi
+
   if [ "$ALEMBIC_SKIP" = false ]; then
     _1="-D WITH_ALEMBIC=ON"
     PRINT "  $_1"

build_files/buildbot/config/blender_linux.cmake

@@ -70,7 +70,7 @@ set(FFMPEG_LIBRARIES
 )
 
 # SndFile libraries
-set(SNDFILE_LIBRARY          "/usr/lib/libsndfile.a;/usr/lib/libFLAC.a" CACHE STRING "" FORCE)
+set(SNDFILE_LIBRARY          "/usr/lib${MULTILIB}/libsndfile.a;/usr/lib${MULTILIB}/libFLAC.a" CACHE STRING "" FORCE)
 
 # OpenAL libraries
 set(OPENAL_ROOT_DIR           "/opt/lib/openal" CACHE STRING "" FORCE)

build_files/buildbot/config/blender_linux_player.cmake

@@ -1,4 +1,4 @@
-# This is applied as an ovveride on top of blender_linux.config
+# This is applied as an override on top of blender_linux.config
 # Disables all the areas which are not needed for the player.
 set(WITH_COMPOSITOR          OFF CACHE BOOL "" FORCE)
 set(WITH_CYCLES              OFF CACHE BOOL "" FORCE)

build_files/buildbot/master_unpack.py

@@ -49,7 +49,7 @@ def get_platform(filename):
     tokens = filename.split("-")
     platforms = ('osx', 'mac', 'bsd',
                  'win', 'linux', 'source',
-                 'irix', 'solaris', 'mingw')
+                 'irix', 'solaris')
     platform_tokens = []
     found = False
 

build_files/buildbot/slave_compile.py

@@ -112,8 +112,8 @@ if 'cmake' in builder:
             chroot_name = 'buildbot_' + deb_name + '_i686'
             cuda_chroot_name = 'buildbot_' + deb_name + '_x86_64'
             targets = ['player', 'blender', 'cuda']
-        cmake_extra_options.extend(["-DCMAKE_C_COMPILER=/usr/bin/gcc-6",
-                                    "-DCMAKE_CXX_COMPILER=/usr/bin/g++-6"])
+        cmake_extra_options.extend(["-DCMAKE_C_COMPILER=/usr/bin/gcc-7",
+                                    "-DCMAKE_CXX_COMPILER=/usr/bin/g++-7"])
 
     cmake_options.append("-C" + os.path.join(blender_dir, cmake_config_file))
 

build_files/cmake/Modules/FindOpenCOLLADA.cmake

@@ -83,6 +83,7 @@ FOREACH(COMPONENT ${_opencollada_FIND_INCLUDES})
       # but this is less trouble, just looks strange.
       include/opencollada/${COMPONENT}
       include/${COMPONENT}/include
+      include/${COMPONENT}
     HINTS
       ${_opencollada_SEARCH_DIRS}
     )

build_files/cmake/cmake_consistency_check.py

@@ -61,10 +61,8 @@ def replace_line(f, i, text, keep_indent=True):
 
 def source_list(path, filename_check=None):
     for dirpath, dirnames, filenames in os.walk(path):
-
         # skip '.git'
-        if dirpath.startswith("."):
-            continue
+        dirnames[:] = [d for d in dirnames if not d.startswith(".")]
 
         for filename in filenames:
             if filename_check is None or filename_check(filename):

build_files/cmake/config/blender_full.cmake

@@ -63,17 +63,10 @@ if(UNIX AND NOT APPLE)
 	set(WITH_OPENSUBDIV          ON  CACHE BOOL "" FORCE)
 elseif(WIN32)
 	set(WITH_JACK                OFF CACHE BOOL "" FORCE)
-	if(NOT CMAKE_COMPILER_IS_GNUCC)
-		set(WITH_OPENSUBDIV          ON  CACHE BOOL "" FORCE)
-	else()
-		# MinGW exceptions
-		set(WITH_OPENSUBDIV          OFF CACHE BOOL "" FORCE)
-		set(WITH_CODEC_SNDFILE       OFF CACHE BOOL "" FORCE)
-		set(WITH_CYCLES_OSL          OFF CACHE BOOL "" FORCE)
-	endif()
+	set(WITH_OPENSUBDIV          ON  CACHE BOOL "" FORCE)
 elseif(APPLE)
 	set(WITH_JACK                ON  CACHE BOOL "" FORCE)
-	set(WITH_CODEC_QUICKTIME     OFF  CACHE BOOL "" FORCE)
+	set(WITH_CODEC_QUICKTIME     OFF CACHE BOOL "" FORCE)
 	set(WITH_OPENSUBDIV          OFF CACHE BOOL "" FORCE)
 
 #	include("${CMAKE_CURRENT_SOURCE_DIR}/../platform/platform_apple_xcode.cmake")

build_files/cmake/config/blender_release.cmake

@@ -64,17 +64,10 @@ if(UNIX AND NOT APPLE)
 	set(WITH_OPENSUBDIV          ON  CACHE BOOL "" FORCE)
 elseif(WIN32)
 	set(WITH_JACK                OFF CACHE BOOL "" FORCE)
-	if(NOT CMAKE_COMPILER_IS_GNUCC)
-		set(WITH_OPENSUBDIV          ON  CACHE BOOL "" FORCE)
-	else()
-		# MinGW exceptions
-		set(WITH_OPENSUBDIV          OFF CACHE BOOL "" FORCE)
-		set(WITH_CODEC_SNDFILE       OFF CACHE BOOL "" FORCE)
-		set(WITH_CYCLES_OSL          OFF CACHE BOOL "" FORCE)
-	endif()
+	set(WITH_OPENSUBDIV          ON  CACHE BOOL "" FORCE)
 elseif(APPLE)
 	set(WITH_JACK                ON  CACHE BOOL "" FORCE)
-	set(WITH_CODEC_QUICKTIME     OFF  CACHE BOOL "" FORCE)
+	set(WITH_CODEC_QUICKTIME     OFF CACHE BOOL "" FORCE)
 	set(WITH_OPENSUBDIV          OFF CACHE BOOL "" FORCE)
 
 #	include("${CMAKE_CURRENT_SOURCE_DIR}/../platform/platform_apple_xcode.cmake")

build_files/cmake/macros.cmake

@@ -1247,17 +1247,6 @@ endfunction()
 # hacks to override initial project settings
 # these macros must be called directly before/after project(Blender)
 macro(blender_project_hack_pre)
-	# ----------------
-	# MINGW HACK START
-	# ignore system set flag, use our own
-	# must be before project(...)
-	# if the user wants to add their own its ok after first run.
-	if(DEFINED CMAKE_C_STANDARD_LIBRARIES)
-		set(_reset_standard_libraries OFF)
-	else()
-		set(_reset_standard_libraries ON)
-	endif()
-
 	# ------------------
 	# GCC -O3 HACK START
 	# needed because O3 can cause problems but
@@ -1276,25 +1265,6 @@ endmacro()
 
 
 macro(blender_project_hack_post)
-	# --------------
-	# MINGW HACK END
-	if(_reset_standard_libraries)
-		# Must come after projecINCt(...)
-		#
-		# MINGW workaround for -ladvapi32 being included which surprisingly causes
-		# string formatting of floats, eg: printf("%.*f", 3, value). to crash blender
-		# with a meaningless stack trace. by overriding this flag we ensure we only
-		# have libs we define.
-		set(CMAKE_C_STANDARD_LIBRARIES "" CACHE STRING "" FORCE)
-		set(CMAKE_CXX_STANDARD_LIBRARIES "" CACHE STRING "" FORCE)
-		mark_as_advanced(
-			CMAKE_C_STANDARD_LIBRARIES
-			CMAKE_CXX_STANDARD_LIBRARIES
-		)
-	endif()
-	unset(_reset_standard_libraries)
-
-
 	# ----------------
 	# GCC -O3 HACK END
 	if(_reset_standard_cflags_rel)

build_files/cmake/platform/platform_win32.cmake

@@ -27,12 +27,10 @@ add_definitions(-DWIN32)
 
 if(MSVC)
 	include(platform_win32_msvc)
-elseif(CMAKE_COMPILER_IS_GNUCC)
-	include(platform_win32_mingw)
+else()
+	message(FATAL_ERROR "Compiler is unsupported")
 endif()
 
-# Things common to both mingw and MSVC  should go here
-
 set(WINTAB_INC ${LIBDIR}/wintab/include)
 
 if(WITH_OPENAL)

build_files/cmake/platform/platform_win32_mingw.cmake

@@ -1,302 +0,0 @@
-# ***** BEGIN GPL LICENSE BLOCK *****
-#
-# This program is free software; you can redistribute it and/or
-# modify it under the terms of the GNU General Public License
-# as published by the Free Software Foundation; either version 2
-# of the License, or (at your option) any later version.
-#
-# This program is distributed in the hope that it will be useful,
-# but WITHOUT ANY WARRANTY; without even the implied warranty of
-# MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the
-# GNU General Public License for more details.
-#
-# You should have received a copy of the GNU General Public License
-# along with this program; if not, write to the Free Software Foundation,
-# Inc., 51 Franklin Street, Fifth Floor, Boston, MA 02110-1301, USA.
-#
-# The Original Code is Copyright (C) 2016, Blender Foundation
-# All rights reserved.
-#
-# Contributor(s): Sergey Sharybin.
-#
-# ***** END GPL LICENSE BLOCK *****
-
-# Libraries configuration for Windows when compiling with MinGW.
-
-# keep GCC specific stuff here
-include(CheckCSourceCompiles)
-# Setup 64bit and 64bit windows systems
-CHECK_C_SOURCE_COMPILES("
-	#ifndef __MINGW64__
-	#error
-	#endif
-	int main(void) { return 0; }
-	"
-	WITH_MINGW64
-)
-
-if(NOT DEFINED LIBDIR)
-	if(WITH_MINGW64)
-		message(STATUS "Compiling for 64 bit with MinGW-w64.")
-		set(LIBDIR ${CMAKE_SOURCE_DIR}/../lib/mingw64)
-	else()
-		message(STATUS "Compiling for 32 bit with MinGW-w32.")
-		set(LIBDIR ${CMAKE_SOURCE_DIR}/../lib/mingw32)
-
-		if(WITH_RAYOPTIMIZATION)
-			message(WARNING "MinGW-w32 is known to be unstable with 'WITH_RAYOPTIMIZATION' option enabled.")
-		endif()
-	endif()
-else()
-	message(STATUS "Using pre-compiled LIBDIR: ${LIBDIR}")
-endif()
-if(NOT EXISTS "${LIBDIR}/")
-	message(FATAL_ERROR "Windows requires pre-compiled libs at: '${LIBDIR}'")
-endif()
-
-list(APPEND PLATFORM_LINKLIBS
-	-lshell32 -lshfolder -lgdi32 -lmsvcrt -lwinmm -lmingw32 -lm -lws2_32
-	-lz -lstdc++ -lole32 -luuid -lwsock32 -lpsapi -ldbghelp
-)
-
-if(WITH_INPUT_IME)
-	list(APPEND PLATFORM_LINKLIBS -limm32)
-endif()
-
-set(PLATFORM_CFLAGS "-pipe -funsigned-char -fno-strict-aliasing")
-
-if(WITH_MINGW64)
-	set(CMAKE_CXX_FLAGS "${CMAKE_CXX_FLAGS} -fpermissive")
-	list(APPEND PLATFORM_LINKLIBS -lpthread)
-
-	add_definitions(-DFREE_WINDOWS64 -DMS_WIN64)
-endif()
-
-add_definitions(-D_LARGEFILE_SOURCE -D_FILE_OFFSET_BITS=64 -D_LARGEFILE64_SOURCE)
-
-add_definitions(-DFREE_WINDOWS)
-
-set(PNG "${LIBDIR}/png")
-set(PNG_INCLUDE_DIRS "${PNG}/include")
-set(PNG_LIBPATH ${PNG}/lib) # not cmake defined
-
-if(WITH_MINGW64)
-	set(JPEG_LIBRARIES jpeg)
-else()
-	set(JPEG_LIBRARIES libjpeg)
-endif()
-set(PNG_LIBRARIES png)
-
-set(ZLIB ${LIBDIR}/zlib)
-set(ZLIB_INCLUDE_DIRS ${ZLIB}/include)
-set(ZLIB_LIBPATH ${ZLIB}/lib)
-set(ZLIB_LIBRARIES z)
-
-set(JPEG "${LIBDIR}/jpeg")
-set(JPEG_INCLUDE_DIR "${JPEG}/include")
-set(JPEG_LIBPATH ${JPEG}/lib) # not cmake defined
-
-# comes with own pthread library
-if(NOT WITH_MINGW64)
-	set(PTHREADS ${LIBDIR}/pthreads)
-	#set(PTHREADS_INCLUDE_DIRS ${PTHREADS}/include)
-	set(PTHREADS_LIBPATH ${PTHREADS}/lib)
-	set(PTHREADS_LIBRARIES pthreadGC2)
-endif()
-
-set(FREETYPE ${LIBDIR}/freetype)
-set(FREETYPE_INCLUDE_DIRS ${FREETYPE}/include ${FREETYPE}/include/freetype2)
-set(FREETYPE_LIBPATH ${FREETYPE}/lib)
-set(FREETYPE_LIBRARY freetype)
-
-if(WITH_FFTW3)
-	set(FFTW3 ${LIBDIR}/fftw3)
-	set(FFTW3_LIBRARIES fftw3)
-	set(FFTW3_INCLUDE_DIRS ${FFTW3}/include)
-	set(FFTW3_LIBPATH ${FFTW3}/lib)
-endif()
-
-if(WITH_OPENCOLLADA)
-	set(OPENCOLLADA ${LIBDIR}/opencollada)
-	set(OPENCOLLADA_INCLUDE_DIRS
-		${OPENCOLLADA}/include/opencollada/COLLADAStreamWriter
-		${OPENCOLLADA}/include/opencollada/COLLADABaseUtils
-		${OPENCOLLADA}/include/opencollada/COLLADAFramework
-		${OPENCOLLADA}/include/opencollada/COLLADASaxFrameworkLoader
-		${OPENCOLLADA}/include/opencollada/GeneratedSaxParser
-	)
-	set(OPENCOLLADA_LIBPATH ${OPENCOLLADA}/lib/opencollada)
-	set(OPENCOLLADA_LIBRARIES
-		OpenCOLLADAStreamWriter
-		OpenCOLLADASaxFrameworkLoader
-		OpenCOLLADAFramework
-		OpenCOLLADABaseUtils
-		GeneratedSaxParser
-		UTF MathMLSolver buffer ftoa xml
-	)
-	set(PCRE_LIBRARIES pcre)
-endif()
-
-if(WITH_CODEC_FFMPEG)
-	set(FFMPEG ${LIBDIR}/ffmpeg)
-	set(FFMPEG_INCLUDE_DIRS ${FFMPEG}/include)
-	if(WITH_MINGW64)
-		set(FFMPEG_LIBRARIES avcodec.dll avformat.dll avdevice.dll avutil.dll swscale.dll swresample.dll)
-	else()
-		set(FFMPEG_LIBRARIES avcodec-55 avformat-55 avdevice-55 avutil-52 swscale-2)
-	endif()
-	set(FFMPEG_LIBPATH ${FFMPEG}/lib)
-endif()
-
-if(WITH_IMAGE_OPENEXR)
-	set(OPENEXR ${LIBDIR}/openexr)
-	set(OPENEXR_INCLUDE_DIR ${OPENEXR}/include)
-	set(OPENEXR_INCLUDE_DIRS ${OPENEXR}/include/OpenEXR)
-	set(OPENEXR_LIBRARIES Half IlmImf Imath IlmThread Iex)
-	set(OPENEXR_LIBPATH ${OPENEXR}/lib)
-endif()
-
-if(WITH_IMAGE_TIFF)
-	set(TIFF ${LIBDIR}/tiff)
-	set(TIFF_LIBRARY tiff)
-	set(TIFF_INCLUDE_DIR ${TIFF}/include)
-	set(TIFF_LIBPATH ${TIFF}/lib)
-endif()
-
-if(WITH_JACK)
-	set(JACK ${LIBDIR}/jack)
-	set(JACK_INCLUDE_DIRS ${JACK}/include/jack ${JACK}/include)
-	set(JACK_LIBRARIES jack)
-	set(JACK_LIBPATH ${JACK}/lib)
-
-	# TODO, gives linking errors, force off
-	set(WITH_JACK OFF)
-endif()
-
-if(WITH_PYTHON)
-	# normally cached but not since we include them with blender
-	set(PYTHON_VERSION 3.5) #  CACHE STRING)
-	string(REPLACE "." "" _PYTHON_VERSION_NO_DOTS ${PYTHON_VERSION})
-	set(PYTHON_INCLUDE_DIR "${LIBDIR}/python/include/python${PYTHON_VERSION}")  # CACHE PATH)
-	set(PYTHON_LIBRARY "${LIBDIR}/python/lib/python${_PYTHON_VERSION_NO_DOTS}mw.lib")  # CACHE FILEPATH)
-	unset(_PYTHON_VERSION_NO_DOTS)
-
-	# uncached vars
-	set(PYTHON_INCLUDE_DIRS "${PYTHON_INCLUDE_DIR}")
-	set(PYTHON_LIBRARIES  "${PYTHON_LIBRARY}")
-endif()
-
-if(WITH_BOOST)
-	set(BOOST ${LIBDIR}/boost)
-	set(BOOST_INCLUDE_DIR ${BOOST}/include)
-	if(WITH_MINGW64)
-		set(BOOST_POSTFIX "mgw47-mt-s-1_49")
-		set(BOOST_DEBUG_POSTFIX "mgw47-mt-sd-1_49")
-	else()
-		set(BOOST_POSTFIX "mgw46-mt-s-1_49")
-		set(BOOST_DEBUG_POSTFIX "mgw46-mt-sd-1_49")
-	endif()
-	set(BOOST_LIBRARIES
-		optimized boost_date_time-${BOOST_POSTFIX} boost_filesystem-${BOOST_POSTFIX}
-		boost_regex-${BOOST_POSTFIX}
-		boost_system-${BOOST_POSTFIX} boost_thread-${BOOST_POSTFIX}
-		debug boost_date_time-${BOOST_DEBUG_POSTFIX} boost_filesystem-${BOOST_DEBUG_POSTFIX}
-		boost_regex-${BOOST_DEBUG_POSTFIX}
-		boost_system-${BOOST_DEBUG_POSTFIX} boost_thread-${BOOST_DEBUG_POSTFIX})
-	if(WITH_INTERNATIONAL)
-		set(BOOST_LIBRARIES ${BOOST_LIBRARIES}
-			optimized boost_locale-${BOOST_POSTFIX}
-			debug boost_locale-${BOOST_DEBUG_POSTFIX}
-		)
-	endif()
-	if(WITH_CYCLES_OSL)
-		set(BOOST_LIBRARIES ${BOOST_LIBRARIES}
-			optimized boost_wave-${BOOST_POSTFIX}
-			debug boost_wave-${BOOST_DEBUG_POSTFIX}
-		)
-	endif()
-	set(BOOST_LIBPATH ${BOOST}/lib)
-	set(BOOST_DEFINITIONS "-DBOOST_ALL_NO_LIB -DBOOST_THREAD_USE_LIB ")
-endif()
-
-if(WITH_OPENIMAGEIO)
-	set(OPENIMAGEIO ${LIBDIR}/openimageio)
-	set(OPENIMAGEIO_INCLUDE_DIRS ${OPENIMAGEIO}/include)
-	set(OPENIMAGEIO_LIBRARIES OpenImageIO)
-	set(OPENIMAGEIO_LIBPATH ${OPENIMAGEIO}/lib)
-	set(OPENIMAGEIO_DEFINITIONS "")
-	set(OPENIMAGEIO_IDIFF "${OPENIMAGEIO}/bin/idiff.exe")
-endif()
-
-if(WITH_LLVM)
-	set(LLVM_ROOT_DIR ${LIBDIR}/llvm CACHE PATH	"Path to the LLVM installation")
-	set(LLVM_LIBPATH ${LLVM_ROOT_DIR}/lib)
-	# Explicitly set llvm lib order.
-	#---- WARNING ON GCC ORDER OF LIBS IS IMPORTANT, DO NOT CHANGE! ---------
-	set(LLVM_LIBRARY LLVMSelectionDAG LLVMCodeGen LLVMScalarOpts LLVMAnalysis LLVMArchive
-		LLVMAsmParser LLVMAsmPrinter
-		LLVMBitReader LLVMBitWriter
-		LLVMDebugInfo LLVMExecutionEngine
-		LLVMInstCombine LLVMInstrumentation
-		LLVMInterpreter LLVMJIT
-		LLVMLinker LLVMMC
-		LLVMMCDisassembler LLVMMCJIT
-		LLVMMCParser LLVMObject
-		LLVMRuntimeDyld
-		LLVMSupport
-		LLVMTableGen LLVMTarget
-		LLVMTransformUtils LLVMVectorize
-		LLVMX86AsmParser LLVMX86AsmPrinter
-		LLVMX86CodeGen LLVMX86Desc
-		LLVMX86Disassembler LLVMX86Info
-		LLVMX86Utils LLVMipa
-		LLVMipo LLVMCore)
-	# imagehelp is needed by LLVM 3.1 on MinGW, check lib\Support\Windows\Signals.inc
-	list(APPEND PLATFORM_LINKLIBS -limagehlp)
-endif()
-
-if(WITH_OPENCOLORIO)
-	set(OPENCOLORIO ${LIBDIR}/opencolorio)
-	set(OPENCOLORIO_INCLUDE_DIRS ${OPENCOLORIO}/include)
-	set(OPENCOLORIO_LIBRARIES OpenColorIO)
-	set(OPENCOLORIO_LIBPATH ${OPENCOLORIO}/lib)
-	set(OPENCOLORIO_DEFINITIONS)
-endif()
-
-if(WITH_SDL)
-	set(SDL ${LIBDIR}/sdl)
-	set(SDL_INCLUDE_DIR ${SDL}/include)
-	set(SDL_LIBRARY SDL)
-	set(SDL_LIBPATH ${SDL}/lib)
-endif()
-
-if(WITH_OPENVDB)
-	set(OPENVDB ${LIBDIR}/openvdb)
-	set(OPENVDB_INCLUDE_DIRS ${OPENVDB}/include)
-	set(OPENVDB_LIBRARIES openvdb ${TBB_LIBRARIES})
-	set(OPENVDB_LIBPATH ${LIBDIR}/openvdb/lib)
-	set(OPENVDB_DEFINITIONS)
-endif()
-
-if(WITH_ALEMBIC)
-	# TODO(sergey): For until someone drops by and compiles libraries for
-	# MinGW we allow users to compile their own Alembic library and use
-	# that via find_package(),
-	#
-	# Once precompiled libraries are there we'll use hardcoded locations.
-	find_package_wrapper(Alembic)
-	if(WITH_ALEMBIC_HDF5)
-		set(HDF5_ROOT_DIR ${LIBDIR}/hdf5)
-		find_package_wrapper(HDF5)
-	endif()
-	if(NOT ALEMBIC_FOUND OR (WITH_ALEMBIC_HDF5 AND NOT HDF5_FOUND))
-		set(WITH_ALEMBIC OFF)
-		set(WITH_ALEMBIC_HDF5 OFF)
-	endif()
-endif()
-
-set(PLATFORM_LINKFLAGS "-Xlinker --stack=2097152")
-
-## DISABLE - causes linking errors
-## for re-distribution, so users dont need mingw installed
-# set(PLATFORM_LINKFLAGS "${PLATFORM_LINKFLAGS} -static-libgcc -static-libstdc++")

build_files/cmake/platform/platform_win32_msvc.cmake

@@ -116,7 +116,6 @@ set(PLATFORM_LINKFLAGS "${PLATFORM_LINKFLAGS} /NODEFAULTLIB:msvcrt.lib /NODEFAUL
 set(PLATFORM_LINKFLAGS "${PLATFORM_LINKFLAGS} /ignore:4049 /ignore:4217 /ignore:4221")
 set(CMAKE_STATIC_LINKER_FLAGS "${CMAKE_STATIC_LINKER_FLAGS} /ignore:4221")
 
-# MSVC only, Mingw doesnt need
 if(CMAKE_CL_64)
 	set(PLATFORM_LINKFLAGS "/MACHINE:X64 ${PLATFORM_LINKFLAGS}")
 else()
@@ -453,32 +452,28 @@ if(WITH_MOD_CLOTH_ELTOPO)
 endif()
 
 if(WITH_OPENSUBDIV OR WITH_CYCLES_OPENSUBDIV)
-    set(OPENSUBDIV_INCLUDE_DIR ${LIBDIR}/opensubdiv/include)
-    set(OPENSUBDIV_LIBPATH ${LIBDIR}/opensubdiv/lib)
-    set(OPENSUBDIV_LIBRARIES    optimized ${OPENSUBDIV_LIBPATH}/osdCPU.lib 
-                                optimized ${OPENSUBDIV_LIBPATH}/osdGPU.lib
-                                debug ${OPENSUBDIV_LIBPATH}/osdCPU_d.lib 
-                                debug ${OPENSUBDIV_LIBPATH}/osdGPU_d.lib
-                                )
-    set(OPENSUBDIV_HAS_OPENMP TRUE)
+	set(OPENSUBDIV_INCLUDE_DIR ${LIBDIR}/opensubdiv/include)
+	set(OPENSUBDIV_LIBPATH ${LIBDIR}/opensubdiv/lib)
+	set(OPENSUBDIV_LIBRARIES
+		optimized ${OPENSUBDIV_LIBPATH}/osdCPU.lib
+		optimized ${OPENSUBDIV_LIBPATH}/osdGPU.lib
+		debug ${OPENSUBDIV_LIBPATH}/osdCPU_d.lib
+		debug ${OPENSUBDIV_LIBPATH}/osdGPU_d.lib
+	)
+	set(OPENSUBDIV_HAS_OPENMP TRUE)
 	set(OPENSUBDIV_HAS_TBB FALSE)
 	set(OPENSUBDIV_HAS_OPENCL TRUE)
 	set(OPENSUBDIV_HAS_CUDA FALSE)
 	set(OPENSUBDIV_HAS_GLSL_TRANSFORM_FEEDBACK TRUE)
 	set(OPENSUBDIV_HAS_GLSL_COMPUTE TRUE)
-    windows_find_package(OpenSubdiv)
+	windows_find_package(OpenSubdiv)
 endif()
 
 if(WITH_SDL)
 	set(SDL ${LIBDIR}/sdl)
 	set(SDL_INCLUDE_DIR ${SDL}/include)
 	set(SDL_LIBPATH ${SDL}/lib)
-	# MinGW TODO: Update MinGW to SDL2
-	if(NOT CMAKE_COMPILER_IS_GNUCC)
-		set(SDL_LIBRARY SDL2)
-	else()
-		set(SDL_LIBRARY SDL)
-	endif()
+	set(SDL_LIBRARY SDL2)
 endif()
 
 # Audio IO
@@ -494,14 +489,14 @@ endif()
 # used in many places so include globally, like OpenGL
 blender_include_dirs_sys("${PTHREADS_INCLUDE_DIRS}")
 
-#find signtool  
-SET(ProgramFilesX86_NAME "ProgramFiles(x86)") #env dislikes the ( ) 
+#find signtool
+set(ProgramFilesX86_NAME "ProgramFiles(x86)") #env dislikes the ( )
 find_program(SIGNTOOL_EXE signtool
-HINTS
-  "$ENV{${ProgramFilesX86_NAME}}/Windows Kits/10/bin/x86/"
-  "$ENV{ProgramFiles}/Windows Kits/10/bin/x86/"
-  "$ENV{${ProgramFilesX86_NAME}}/Windows Kits/8.1/bin/x86/"
-  "$ENV{ProgramFiles}/Windows Kits/8.1/bin/x86/"
-  "$ENV{${ProgramFilesX86_NAME}}/Windows Kits/8.0/bin/x86/"
-  "$ENV{ProgramFiles}/Windows Kits/8.0/bin/x86/"
+	HINTS
+		"$ENV{${ProgramFilesX86_NAME}}/Windows Kits/10/bin/x86/"
+		"$ENV{ProgramFiles}/Windows Kits/10/bin/x86/"
+		"$ENV{${ProgramFilesX86_NAME}}/Windows Kits/8.1/bin/x86/"
+		"$ENV{ProgramFiles}/Windows Kits/8.1/bin/x86/"
+		"$ENV{${ProgramFilesX86_NAME}}/Windows Kits/8.0/bin/x86/"
+		"$ENV{ProgramFiles}/Windows Kits/8.0/bin/x86/"
 )

build_files/cmake/project_info.py

@@ -84,10 +84,8 @@ def init(cmake_path):
 
 def source_list(path, filename_check=None):
     for dirpath, dirnames, filenames in os.walk(path):
-
-        # skip '.svn'
-        if dirpath.startswith("."):
-            continue
+        # skip '.git'
+        dirnames[:] = [d for d in dirnames if not d.startswith(".")]
 
         for filename in filenames:
             filepath = join(dirpath, filename)
@@ -145,7 +143,7 @@ def cmake_advanced_info():
     def create_eclipse_project():
         print("CMAKE_DIR %r" % CMAKE_DIR)
         if sys.platform == "win32":
-            cmd = 'cmake "%s" -G"Eclipse CDT4 - MinGW Makefiles"' % CMAKE_DIR
+            raise Exception("Error: win32 is not supported")
         else:
             if make_exe_basename.startswith(("make", "gmake")):
                 cmd = 'cmake "%s" -G"Eclipse CDT4 - Unix Makefiles"' % CMAKE_DIR

doc/doxygen/Doxyfile

@@ -38,7 +38,7 @@ PROJECT_NAME           = Blender
 # could be handy for archiving the generated documentation or if some version
 # control system is used.
 
-PROJECT_NUMBER         = "V2.8x"
+PROJECT_NUMBER         = "V2.79"
 
 # Using the PROJECT_BRIEF tag one can provide an optional one line description
 # for a project that appears at the top of each page and should give viewer a

doc/python_api/examples/bpy.app.translations.py

@@ -32,3 +32,62 @@ Module References
 -----------------
 
 """
+
+import bpy
+
+# This block can be automatically generated by UI translations addon, which also handles conversion with PO format.
+# See also https://wiki.blender.org/index.php/Dev:Doc/Process/Translate_Blender#Translating_non-official_addons
+# It can (should) also be put in a different, specific py file.
+
+# ##### BEGIN AUTOGENERATED I18N SECTION #####
+# NOTE: You can safely move around this auto-generated block (with the begin/end markers!),
+#       and edit the translations by hand.
+#       Just carefully respect the format of the tuple!
+
+# Tuple of tuples ((msgctxt, msgid), (sources, gen_comments), (lang, translation, (is_fuzzy, comments)), ...)
+translations_tuple = (
+    (("*", ""),
+     ((), ()),
+     ("fr_FR", "Project-Id-Version: Copy Settings 0.1.5 (r0)\nReport-Msgid-Bugs-To: \nPOT-Creation-Date: 2013-04-18 15:27:45.563524\nPO-Revision-Date: 2013-04-18 15:38+0100\nLast-Translator: Bastien Montagne <montagne29@wanadoo.fr>\nLanguage-Team: LANGUAGE <LL@li.org>\nLanguage: __POT__\nMIME-Version: 1.0\nContent-Type: text/plain; charset=UTF-8\nContent-Transfer-Encoding: 8bit\n",
+               (False,
+                ("Blender's translation file (po format).",
+                 "Copyright (C) 2013 The Blender Foundation.",
+                 "This file is distributed under the same license as the Blender package.",
+                 "FIRST AUTHOR <EMAIL@ADDRESS>, YEAR."))),
+    ),
+    (("Operator", "Render: Copy Settings"),
+     (("bpy.types.SCENE_OT_render_copy_settings",),
+      ()),
+     ("fr_FR", "Rendu : copier réglages",
+               (False, ())),
+    ),
+    (("*", "Copy render settings from current scene to others"),
+     (("bpy.types.SCENE_OT_render_copy_settings",),
+      ()),
+     ("fr_FR", "Copier les réglages de rendu depuis la scène courante vers d’autres",
+               (False, ())),
+    ),
+    # ... etc, all messages from your addon.
+)
+
+translations_dict = {}
+for msg in translations_tuple:
+    key = msg[0]
+    for lang, trans, (is_fuzzy, comments) in msg[2:]:
+        if trans and not is_fuzzy:
+            translations_dict.setdefault(lang, {})[key] = trans
+
+# ##### END AUTOGENERATED I18N SECTION #####
+
+# Define remaining addon (operators, UI...) here.
+
+def register():
+   # Usual operator/UI/etc. registration...
+
+    bpy.app.translations.register(__name__, translations_dict)
+
+
+def unregister():
+    bpy.app.translations.unregister(__name__)
+
+   # Usual operator/UI/etc. unregistration...

doc/python_api/examples/bpy.types.Operator.5.py

@@ -2,13 +2,14 @@
 Modal Execution
 +++++++++++++++
 
-This operator defines a :class:`Operator.modal` function which running,
-handling events until it returns ``{'FINISHED'}`` or ``{'CANCELLED'}``.
+This operator defines a :class:`Operator.modal` function that will keep being
+run to handle events until it returns ``{'FINISHED'}`` or ``{'CANCELLED'}``.
 
-Grab, Rotate, Scale and Fly-Mode are examples of modal operators.
-They are especially useful for interactive tools,
-your operator can have its own state where keys toggle options as the operator
-runs.
+Modal operators run every time a new event is detected, such as a mouse click
+or key press. Conversely, when no new events are detected, the modal operator
+will not run. Modal operators are especially useful for interactive tools, an
+operator can have its own state where keys toggle options as the operator runs.
+Grab, Rotate, Scale, and Fly-Mode are examples of modal operators.
 
 :class:`Operator.invoke` is used to initialize the operator as being by
 returning ``{'RUNNING_MODAL'}``, initializing the modal loop.

doc/python_api/rst/bge.app.rst

@@ -47,4 +47,3 @@ Module to access application values that remain unchanged during runtime.
    True if the BGE has been built with physics support.
 
    :type: bool
-

doc/python_api/rst/bge_types/bge.types.BL_ShapeActionActuator.rst

@@ -59,7 +59,7 @@ base class --- :class:`SCA_IActuator`
 
    .. attribute:: mode
 
-      The operation mode of the actuator. Can be one of :ref:`these constants<shape-action-actuator>`.
+      The operation mode of the actuator. Can be one of :ref:`these constants<action-actuator>`.
 
       :type: integer
 
@@ -68,5 +68,3 @@ base class --- :class:`SCA_IActuator`
       The name of the property that is set to the current frame number.
 
       :type: string
-
-

doc/python_api/rst/gpu.rst

@@ -6,6 +6,13 @@ GPU functions (gpu)
 
 This module provides access to materials GLSL shaders.
 
+Submodules:
+
+.. toctree::
+   :maxdepth: 1
+
+   gpu.offscreen.rst
+
 
 Intro
 =====
@@ -24,7 +31,6 @@ and in the game engine.
 Constants
 =========
 
-
 GLSL Data Type
 --------------
 

doc/python_api/rst/include__bmesh.rst

@@ -7,14 +7,17 @@
 
 Submodules:
 
-* :mod:`bmesh.ops`
-* :mod:`bmesh.types`
-* :mod:`bmesh.utils`
-* :mod:`bmesh.geometry`
+.. toctree::
+   :maxdepth: 1
+
+   bmesh.ops.rst
+   bmesh.types.rst
+   bmesh.utils.rst
+   bmesh.geometry.rst
 
 
-Intro
------
+Introduction
+------------
 
 This API gives access the blenders internal mesh editing api, featuring geometry connectivity data and
 access to editing operations such as split, separate, collapse and dissolve.

doc/python_api/rst/info_overview.rst

@@ -5,23 +5,25 @@
 Python API Overview
 *******************
 
-This document is to give an understanding of how Python and Blender fit together,
-covering some of the functionality that isn't obvious from reading the API reference and example scripts.
+The purpose of this document is to explain how Python and Blender fit together,
+covering some of the functionality that may not be obvious from reading the API
+references and example scripts.
 
 
 Python in Blender
 =================
 
-Blender embeds a Python interpreter which is started with Blender and stays active.
-This interpreter runs scripts to draw the user interface and is used for some of Blender's internal tools too.
+Blender has an embedded Python interpreter which is loaded when Blender is started and stays
+active while Blender is running. This interpreter runs scripts to draw the user interface
+and is used for some of Blender’s internal tools as well.
 
-This is a typical Python environment so tutorials on how to write Python scripts
-will work running the scripts in Blender too.
-Blender provides the :mod:`bpy` module to the Python interpreter.
-This module can be imported in a script and gives access to Blender data, classes, and functions.
-Scripts that deal with Blender data will need to import this module.
+Blender's embedded interpreter provides a typical Python environment, so code from tutorials
+on how to write Python scripts can also be run with Blender’s interpreter. Blender provides its
+Python modules, such as :mod:`bpy` and :mod:`mathutils`, to the embedded interpreter so they can
+be imported into a script and give access to Blender's data, classes, and functions. Scripts that
+deal with Blender data will need to import the modules to work.
 
-Here is a simple example of moving a vertex of the object named **Cube**:
+Here is a simple example which moves a vertex attached to an object named **Cube**:
 
 .. code-block:: python
 
@@ -49,15 +51,17 @@ See the :ref:`directory layout docs <blender_manual:getting-started_installing-c
 Script Loading
 ==============
 
-This may seem obvious but it's important to note the difference
-between executing a script directly or importing it as a module.
+This may seem obvious, but it is important to note the difference between
+executing a script directly and importing a script as a module.
 
-Scripts that extend Blender - define classes that exist beyond the scripts execution,
-this makes future access to these classes (to unregister for example)
-more difficult than importing as a module where class instance is kept
-in the module and can be accessed by importing that module later on.
+Extending Blender by executing a script directly means the classes that the script
+defines remain available inside Blender after the script finishes execution.
+Using scripts this way makes future access to their classes
+(to unregister them for example) more difficult compared to importing the scripts as modules.
+When a script is imported as a module, its class instances will remain
+inside the module and can be accessed later on by importing that module again.
 
-For this reason it's preferable to only use directly execute scripts that don't extend Blender by registering classes.
+For this reason it is preferable to avoid directly executing scripts that extend Blender by registering classes.
 
 
 Here are some ways to run scripts directly in Blender.
@@ -396,8 +400,8 @@ This works just as well for PropertyGroup subclasses you define yourself.
 Dynamic Defined-Classes (Advanced)
 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
 
-In some cases the specifier for data may not be in Blender,
-renderman shader definitions for example and it may be useful to define types and remove them on the fly.
+In some cases the specifier for data may not be in Blender, renderman shader definitions
+for example, and it may be useful to define them as types and remove them on the fly.
 
 .. code-block:: python
 
@@ -420,7 +424,7 @@ renderman shader definitions for example and it may be useful to define types an
    This is an alternative syntax for class creation in Python, better suited to constructing classes dynamically.
 
 
-Calling these operators:
+To call the operators from the previous example:
 
    >>> bpy.ops.object.operator_1()
    Hello World OBJECT_OT_operator_1

doc/python_api/rst/info_tutorial_addon.rst

@@ -1,635 +0,0 @@
-
-Add-on Tutorial
-###############
-
-************
-Introduction
-************
-
-
-Intended Audience
-=================
-
-This tutorial is designed to help technical artists or developers learn to extend Blender.
-An understanding of the basics of Python is expected for those working through this tutorial.
-
-
-Prerequisites
--------------
-
-Before going through the tutorial you should...
-
-- Familiarity with the basics of working in Blender.
-- Know how to run a script in Blender's text editor (as documented in the quick-start)
-- Have an understanding of Python primitive types (int, boolean, string, list, tuple, dictionary, and set).
-- Be familiar with the concept of Python modules.
-- Basic understanding of classes (object orientation) in Python.
-
-
-Suggested reading before starting this tutorial.
-
-- `Dive Into Python <http://getpython3.com/diveintopython3/index.html>`_ sections (1, 2, 3, 4, and 7).
-- :ref:`Blender API Quickstart <info_quickstart>`
-  to help become familiar with Blender/Python basics.
-
-
-To best troubleshoot any error message Python prints while writing scripts you run blender with from a terminal,
-see :ref:`Use The Terminal <use_the_terminal>`.
-
-
-Documentation Links
-===================
-
-While going through the tutorial you may want to look into our reference documentation.
-
-- :ref:`Blender API Overview <info_overview>`. -
-  *This document is rather detailed but helpful if you want to know more on a topic.*
-- :mod:`bpy.context` api reference. -
-  *Handy to have a list of available items your script may operate on.*
-- :class:`bpy.types.Operator`. -
-  *The following add-ons define operators, these docs give details and more examples of operators.*
-
-
-*******
-Add-ons
-*******
-
-What is an Add-on?
-==================
-
-An add-on is simply a Python module with some additional requirements so Blender can display it in a list with useful
-information.
-
-To give an example, here is the simplest possible add-on.
-
-.. code-block:: python
-
-   bl_info = {"name": "My Test Add-on", "category": "Object"}
-   def register():
-       print("Hello World")
-   def unregister():
-       print("Goodbye World")
-
-
-- ``bl_info`` is a dictionary containing add-on metadata such as the title,
-  version and author to be displayed in the user preferences add-on list.
-- ``register`` is a function which only runs when enabling the add-on,
-  this means the module can be loaded without activating the add-on.
-- ``unregister`` is a function to unload anything setup by ``register``, this is called when the add-on is disabled.
-
-
-Notice this add-on does not do anything related to Blender, (the :mod:`bpy` module is not imported for example).
-
-This is a contrived example of an add-on that serves to illustrate the point
-that the base requirements of an add-on are simple.
-
-An add-on will typically register operators, panels, menu items etc, but its worth noting that _any_ script can do this,
-when executed from the text editor or even the interactive console - there is nothing inherently different about an
-add-on that allows it to integrate with Blender, such functionality is just provided by the :mod:`bpy` module for any
-script to access.
-
-So an add-on is just a way to encapsulate a Python module in a way a user can easily utilize.
-
-.. note::
-
-   Running this script within the text editor won't print anything,
-   to see the output it must be installed through the user preferences.
-   Messages will be printed when enabling and disabling.
-
-
-Your First Add-on
-=================
-
-The simplest possible add-on above is useful as an example but not much else.
-This next add-on is simple but shows how to integrate a script into Blender using an ``Operator``
-which is the typical way to define a tool accessed from menus, buttons and keyboard shortcuts.
-
-For the first example we will make a script that simply moves all objects in a scene.
-
-
-Write The Script
-----------------
-
-Add the following script to the text editor in Blender.
-
-.. code-block:: python
-
-   import bpy
-
-   scene = bpy.context.scene
-   for obj in scene.objects:
-       obj.location.x += 1.0
-
-
-Click the :ref:`Run Script button <blender_manual:editors-text-run-script>`,
-all objects in the active scene are moved by 1.0 Blender unit.
-
-
-Write the Add-on (Simple)
--------------------------
-
-This add-on takes the body of the script above, and adds them to an operator's ``execute()`` function.
-
-
-.. code-block:: python
-
-   bl_info = {
-       "name": "Move X Axis",
-       "category": "Object",
-   }
-
-   import bpy
-
-
-   class ObjectMoveX(bpy.types.Operator):
-       """My Object Moving Script"""      # blender will use this as a tooltip for menu items and buttons.
-       bl_idname = "object.move_x"        # unique identifier for buttons and menu items to reference.
-       bl_label = "Move X by One"         # display name in the interface.
-       bl_options = {'REGISTER', 'UNDO'}  # enable undo for the operator.
-
-       def execute(self, context):        # execute() is called by blender when running the operator.
-
-           # The original script
-           scene = context.scene
-           for obj in scene.objects:
-               obj.location.x += 1.0
-
-           return {'FINISHED'}            # this lets blender know the operator finished successfully.
-
-   def register():
-       bpy.utils.register_class(ObjectMoveX)
-
-
-   def unregister():
-       bpy.utils.unregister_class(ObjectMoveX)
-
-
-   # This allows you to run the script directly from blenders text editor
-   # to test the add-on without having to install it.
-   if __name__ == "__main__":
-       register()
-
-
-.. note::
-
-   ``bl_info`` is split across multiple lines, this is just a style convention used to more easily add items.
-
-.. note::
-
-   Rather than using ``bpy.context.scene``, we use the ``context.scene`` argument passed to ``execute()``.
-   In most cases these will be the same however in some cases operators will be passed a custom context
-   so script authors should prefer the ``context`` argument passed to operators.
-
-To test the script you can copy and paste this into Blender text editor and run it, this will execute the script
-directly and call register immediately.
-
-However running the script wont move any objects, for this you need to execute the newly registered operator.
-
-.. image:: spacebar.png
-   :width: 924px
-   :align: center
-   :height: 574px
-   :alt: Spacebar
-
-Do this by pressing :kbd:`Spacebar` to bring up the operator search dialog and type in
-"Move X by One" (the ``bl_label``), then :kbd:`Enter`.
-
-
-
-The objects should move as before.
-
-*Keep this add-on open in Blender for the next step - Installing.*
-
-
-Install The Add-on
-------------------
-
-Once you have your add-on within in Blender's text editor,
-you will want to be able to install it so it can be enabled in the user preferences to load on startup.
-
-Even though the add-on above is a test, lets go through the steps anyway so you know how to do it for later.
-
-To install the Blender text as an add-on you will first have to save it to disk, take care to obey the naming
-restrictions that apply to Python modules and end with a ``.py`` extension.
-
-Once the file is on disk, you can install it as you would for an add-on downloaded online.
-
-Open the user :menuselection:`File --> User Preferences`,
-Select the *Add-on* section, press *Install Add-on...* and select the file. 
-
-Now the add-on will be listed and you can enable it by pressing the check-box,
-if you want it to be enabled on restart, press *Save as Default*.
-
-.. note::
-
-   The destination of the add-on depends on your Blender configuration.
-   When installing an add-on the source and destination path are printed in the console.
-   You can also find add-on path locations by running this in the Python console.
-
-   .. code-block:: python
-
-      import addon_utils
-      print(addon_utils.paths())
-
-   More is written on this topic here:
-   :ref:`Directory Layout <blender_manual:getting-started_installing-config-directories>`.
-
-
-Your Second Add-on
-==================
-
-For our second add-on, we will focus on object instancing - this is - to make linked copies of an object in a
-similar way to what you may have seen with the array modifier.
-
-
-Write The Script
-----------------
-
-As before, first we will start with a script, develop it, then convert into an add-on.
-
-.. code-block:: python
-
-   import bpy
-   from bpy import context
-
-   # Get the current scene
-   scene = context.scene
-
-   # Get the 3D cursor
-   cursor = scene.cursor_location
-
-   # Get the active object (assume we have one)
-   obj = scene.objects.active
-
-   # Now make a copy of the object
-   obj_new = obj.copy()
-
-   # The object won't automatically get into a new scene
-   scene.objects.link(obj_new)
-
-   # Now we can place the object
-   obj_new.location = cursor
-
-
-Now try copy this script into Blender and run it on the default cube.
-Make sure you click to move the 3D cursor before running as the duplicate will appear at the cursor's location.
-
-
-... go off and test ...
-
-
-After running, notice that when you go into edit-mode to change the cube - all of the copies change,
-in Blender this is known as *Linked-Duplicates*.
-
-
-Next, we're going to do this in a loop, to make an array of objects between the active object and the cursor.
-
-
-.. code-block:: python
-
-   import bpy
-   from bpy import context
-
-   scene = context.scene
-   cursor = scene.cursor_location
-   obj = scene.objects.active
-
-   # Use a fixed value for now, eventually make this user adjustable
-   total = 10
-
-   # Add 'total' objects into the scene
-   for i in range(total):
-       obj_new = obj.copy()
-       scene.objects.link(obj_new)
-
-       # Now place the object in between the cursor
-       # and the active object based on 'i'
-       factor = i / total
-       obj_new.location = (obj.location * factor) + (cursor * (1.0 - factor))
-
-
-Try run this script with with the active object and the cursor spaced apart to see the result.
-
-With this script you'll notice we're doing some math with the object location and cursor, this works because both are
-3D :class:`mathutils.Vector` instances, a convenient class provided by the :mod:`mathutils` module and
-allows vectors to be multiplied by numbers and matrices.
-
-If you are interested in this area, read into :class:`mathutils.Vector` - there are many handy utility functions
-such as getting the angle between vectors, cross product, dot products
-as well as more advanced functions in :mod:`mathutils.geometry` such as Bézier Spline interpolation and
-ray-triangle intersection.
-
-For now we will focus on making this script an add-on, but its good to know that this 3D math module is available and
-can help you with more advanced functionality later on.
-
-
-Write the Add-on
-----------------
-
-The first step is to convert the script as-is into an add-on.
-
-
-.. code-block:: python
-
-   bl_info = {
-       "name": "Cursor Array",
-       "category": "Object",
-   }
-
-   import bpy
-
-
-   class ObjectCursorArray(bpy.types.Operator):
-       """Object Cursor Array"""
-       bl_idname = "object.cursor_array"
-       bl_label = "Cursor Array"
-       bl_options = {'REGISTER', 'UNDO'}
-
-       def execute(self, context):
-           scene = context.scene
-           cursor = scene.cursor_location
-           obj = scene.objects.active
-
-           total = 10
-
-           for i in range(total):
-               obj_new = obj.copy()
-               scene.objects.link(obj_new)
-
-               factor = i / total
-               obj_new.location = (obj.location * factor) + (cursor * (1.0 - factor))
-
-           return {'FINISHED'}
-
-   def register():
-       bpy.utils.register_class(ObjectCursorArray)
-
-
-   def unregister():
-       bpy.utils.unregister_class(ObjectCursorArray)
-
-
-   if __name__ == "__main__":
-       register()
-
-
-Everything here has been covered in the previous steps, you may want to try run the add-on still
-and consider what could be done to make it more useful.
-
-
-... go off and test ...
-
-
-The two of the most obvious missing things are - having the total fixed at 10, and having to access the operator from
-space-bar is not very convenient.
-
-Both these additions are explained next, with the final script afterwards.
-
-
-Operator Property
-^^^^^^^^^^^^^^^^^
-
-There are a variety of property types that are used for tool settings, common property types include:
-int, float, vector, color, boolean and string.
-
-These properties are handled differently to typical Python class attributes
-because Blender needs to be display them in the interface,
-store their settings in key-maps and keep settings for re-use.
-
-While this is handled in a fairly Pythonic way, be mindful that you are in fact defining tool settings that
-are loaded into Blender and accessed by other parts of Blender, outside of Python.
-
-
-To get rid of the literal 10 for `total`, we'll us an operator property.
-Operator properties are defined via bpy.props module, this is added to the class body.
-
-.. code-block:: python
-
-   # moved assignment from execute() to the body of the class...
-   total = bpy.props.IntProperty(name="Steps", default=2, min=1, max=100)
-
-   # and this is accessed on the class
-   # instance within the execute() function as...
-   self.total
-
-
-These properties from :mod:`bpy.props` are handled specially by Blender when the class is registered
-so they display as buttons in the user interface.
-There are many arguments you can pass to properties to set limits, change the default and display a tooltip.
-
-.. seealso:: :mod:`bpy.props.IntProperty`
-
-This document doesn't go into details about using other property types,
-however the link above includes examples of more advanced property usage.
-
-
-Menu Item
-^^^^^^^^^
-
-Add-ons can add to the user interface of existing panels, headers and menus defined in Python.
-
-For this example we'll add to an existing menu.
-
-.. image:: menu_id.png
-   :width: 334px
-   :align: center
-   :height: 128px
-   :alt: Menu Identifier
-
-To find the identifier of a menu you can hover your mouse over the menu item and the identifier is displayed.
-
-The method used for adding a menu item is to append a draw function into an existing class.
-
-
-.. code-block:: python
-
-   def menu_func(self, context):
-       self.layout.operator(ObjectCursorArray.bl_idname)
-
-   def register():
-       bpy.types.VIEW3D_MT_object.append(menu_func)
-
-
-For docs on extending menus see: :doc:`bpy.types.Menu`.
-
-
-Keymap
-^^^^^^
-
-In Blender, add-ons have their own keymaps so as not to interfere with Blenders built in key-maps.
-
-In the example below, a new object-mode :class:`bpy.types.KeyMap` is added,
-then a :class:`bpy.types.KeyMapItem` is added to the key-map which references our newly added operator,
-using :kbd:`Ctrl-Shift-Space` as the key shortcut to activate it.
-
-
-.. code-block:: python
-
-   # store keymaps here to access after registration
-   addon_keymaps = []
-
-   def register():
-
-       # handle the keymap
-       wm = bpy.context.window_manager
-       km = wm.keyconfigs.addon.keymaps.new(name='Object Mode', space_type='EMPTY')
-
-       kmi = km.keymap_items.new(ObjectCursorArray.bl_idname, 'SPACE', 'PRESS', ctrl=True, shift=True)
-       kmi.properties.total = 4
-
-       addon_keymaps.append((km, kmi))
-
-
-   def unregister():
-
-       # handle the keymap
-       for km, kmi in addon_keymaps:
-           km.keymap_items.remove(kmi)
-       addon_keymaps.clear()
-
-
-Notice how the key-map item can have a different ``total`` setting then the default set by the operator,
-this allows you to have multiple keys accessing the same operator with different settings.
-
-
-.. note::
-
-   While :kbd:`Ctrl-Shift-Space` isn't a default Blender key shortcut, its hard to make sure add-ons won't
-   overwrite each others keymaps, At least take care when assigning keys that they don't
-   conflict with important functionality within Blender.
-
-For API documentation on the functions listed above, see:
-:class:`bpy.types.KeyMaps.new`,
-:class:`bpy.types.KeyMap`,
-:class:`bpy.types.KeyMapItems.new`,
-:class:`bpy.types.KeyMapItem`.
-
-
-Bringing it all together
-^^^^^^^^^^^^^^^^^^^^^^^^
-
-.. code-block:: python
-
-   bl_info = {
-       "name": "Cursor Array",
-       "category": "Object",
-   }
-
-   import bpy
-
-
-   class ObjectCursorArray(bpy.types.Operator):
-       """Object Cursor Array"""
-       bl_idname = "object.cursor_array"
-       bl_label = "Cursor Array"
-       bl_options = {'REGISTER', 'UNDO'}
-
-       total = bpy.props.IntProperty(name="Steps", default=2, min=1, max=100)
-
-       def execute(self, context):
-           scene = context.scene
-           cursor = scene.cursor_location
-           obj = scene.objects.active
-
-           for i in range(self.total):
-               obj_new = obj.copy()
-               scene.objects.link(obj_new)
-
-               factor = i / self.total
-               obj_new.location = (obj.location * factor) + (cursor * (1.0 - factor))
-
-           return {'FINISHED'}
-
-
-   def menu_func(self, context):
-       self.layout.operator(ObjectCursorArray.bl_idname)
-
-   # store keymaps here to access after registration
-   addon_keymaps = []
-
-
-   def register():
-       bpy.utils.register_class(ObjectCursorArray)
-       bpy.types.VIEW3D_MT_object.append(menu_func)
-
-       # handle the keymap
-       wm = bpy.context.window_manager
-       # Note that in background mode (no GUI available), keyconfigs are not available either, so we have to check this
-       # to avoid nasty errors in background case.
-       kc = wm.keyconfigs.addon
-       if kc:
-           km = wm.keyconfigs.addon.keymaps.new(name='Object Mode', space_type='EMPTY')
-           kmi = km.keymap_items.new(ObjectCursorArray.bl_idname, 'SPACE', 'PRESS', ctrl=True, shift=True)
-           kmi.properties.total = 4
-           addon_keymaps.append((km, kmi))
-
-   def unregister():
-       # Note: when unregistering, it's usually good practice to do it in reverse order you registered.
-       # Can avoid strange issues like keymap still referring to operators already unregistered...
-       # handle the keymap
-       for km, kmi in addon_keymaps:
-           km.keymap_items.remove(kmi)
-       addon_keymaps.clear()
-
-       bpy.utils.unregister_class(ObjectCursorArray)
-       bpy.types.VIEW3D_MT_object.remove(menu_func)
-
-
-   if __name__ == "__main__":
-       register()
-
-.. image:: in_menu.png
-   :width: 591px
-   :align: center
-   :height: 649px
-   :alt: In the menu
-
-Run the script (or save it and add it through the Preferences like before) and it will appear in the menu.
-
-.. image:: op_prop.png
-   :width: 669px
-   :align: center
-   :height: 644px
-   :alt: Operator Property
-
-After selecting it from the menu, you can choose how many instance of the cube you want created.
-
-
-.. note::
-
-   Directly executing the script multiple times will add the menu each time too.
-   While not useful behavior, theres nothing to worry about since add-ons won't register them selves multiple
-   times when enabled through the user preferences.
-
-
-Conclusions
-===========
-
-Add-ons can encapsulate certain functionality neatly for writing tools to improve your work-flow or for writing utilities
-for others to use.
-
-While there are limits to what Python can do within Blender, there is certainly a lot that can be achieved without
-having to dive into Blender's C/C++ code.
-
-The example given in the tutorial is limited, but shows the Blender API used for common tasks that you can expand on
-to write your own tools.
-
-
-Further Reading
----------------
-
-Blender comes commented templates which are accessible from the text editor header, if you have specific areas
-you want to see example code for, this is a good place to start.
-
-
-Here are some sites you might like to check on after completing this tutorial.
-
-- :ref:`Blender/Python API Overview <info_overview>` -
-  *For more background details on Blender/Python integration.*
-- `How to Think Like a Computer Scientist <http://interactivepython.org/courselib/static/thinkcspy/index.html>`_ -
-  *Great info for those who are still learning Python.*
-- `Blender Development (Wiki) <https://wiki.blender.org/index.php/Dev:Contents>`_ -
-  *Blender Development, general information and helpful links.*
-- `Blender Artists (Coding Section) <https://blenderartists.org/forum/forumdisplay.php?47-Coding>`_ -
-  *forum where people ask Python development questions*
-

doc/python_api/rst_from_bmesh_opdefines.py

@@ -22,7 +22,7 @@
 # 'bmesh_opdefines.c' in order to avoid having to add a lot of introspection
 # data access into the api.
 #
-# The script is stupid becase it makes assumptions about formatting...
+# The script is stupid because it makes assumptions about formatting...
 # that each arg has its own line, that comments above or directly after will be __doc__ etc...
 #
 # We may want to replace this script with something else one day but for now its good enough.

doc/python_api/sphinx_doc_gen.py

@@ -341,6 +341,8 @@ EXTRA_SOURCE_FILES = (
     "../examples/bge.texture.py",
     "../examples/bmesh.ops.1.py",
     "../examples/bpy.app.translations.py",
+    "../static/favicon.ico",
+    "../static/blender_logo.svg",
 )
 
 
@@ -362,8 +364,6 @@ INFO_DOCS = (
      "Blender/Python Quickstart: new to Blender/scripting and want to get your feet wet?"),
     ("info_overview.rst",
      "Blender/Python API Overview: a more complete explanation of Python integration"),
-    ("info_tutorial_addon.rst",
-     "Blender/Python Add-on Tutorial: a step by step guide on how to write an add-on from scratch"),
     ("info_api_reference.rst",
      "Blender/Python API Reference Usage: examples of how to use the API reference docs"),
     ("info_best_practice.rst",
@@ -1613,10 +1613,8 @@ def pyrna2sphinx(basepath):
                     else:
                         url_base = API_BASEURL
 
-                    fw("   :file: `%s <%s/%s>`_:%d\n\n" % (location[0],
-                                                           url_base,
-                                                           location[0],
-                                                           location[1]))
+                    fw("   :file: `%s\\:%d <%s/%s$%d>`_\n\n" %
+                       (location[0], location[1], url_base, location[0], location[1]))
 
             file.close()
 
@@ -1641,19 +1639,24 @@ def write_sphinx_conf_py(basepath):
     fw("version = '%s - API'\n" % BLENDER_VERSION_DOTS)
     fw("release = '%s - API'\n" % BLENDER_VERSION_DOTS)
 
+    # Quiet file not in table-of-contents warnings.
+    fw("exclude_patterns = [\n")
+    fw("    'include__bmesh.rst',\n")
+    fw("]\n\n")
+
     if ARGS.sphinx_theme != 'default':
         fw("html_theme = '%s'\n" % ARGS.sphinx_theme)
 
     if ARGS.sphinx_theme == "blender-org":
         fw("html_theme_path = ['../']\n")
-        # copied with the theme, exclude else we get an error [T28873]
-        fw("html_favicon = 'favicon.ico'\n")    # in <theme>/static/
 
     # not helpful since the source is generated, adds to upload size.
     fw("html_copy_source = False\n")
     fw("html_show_sphinx = False\n")
     fw("html_split_index = True\n")
-    fw("\n")
+    fw("html_extra_path = ['__/static/favicon.ico', '__/static/blender_logo.svg']\n")
+    fw("html_favicon = '__/static/favicon.ico'\n")
+    fw("html_logo = '__/static/blender_logo.svg'\n\n")
 
     # needed for latex, pdf gen
     fw("latex_elements = {\n")
@@ -1661,6 +1664,24 @@ def write_sphinx_conf_py(basepath):
     fw("}\n\n")
 
     fw("latex_documents = [ ('contents', 'contents.tex', 'Blender Index', 'Blender Foundation', 'manual'), ]\n")
+
+    # Workaround for useless links leading to compile errors
+    # See https://github.com/sphinx-doc/sphinx/issues/3866
+    fw(r"""
+from sphinx.domains.python import PythonDomain
+
+class PatchedPythonDomain(PythonDomain):
+    def resolve_xref(self, env, fromdocname, builder, typ, target, node, contnode):
+        if 'refspecific' in node:
+            del node['refspecific']
+        return super(PatchedPythonDomain, self).resolve_xref(
+            env, fromdocname, builder, typ, target, node, contnode)
+
+def setup(sphinx):
+    sphinx.override_domain(PatchedPythonDomain)
+""")
+    # end workaround
+
     file.close()
 
 
@@ -1715,8 +1736,6 @@ def write_rst_contents(basepath):
         "bpy.utils.previews",
         "bpy.path",
         "bpy.app",
-        "bpy.app.handlers",
-        "bpy.app.translations",
 
         # C modules
         "bpy.props",
@@ -1731,19 +1750,9 @@ def write_rst_contents(basepath):
     fw("   :maxdepth: 1\n\n")
 
     standalone_modules = (
-        # mathutils
-        "mathutils",
-        "mathutils.geometry",
-        "mathutils.bvhtree", "mathutils.kdtree",
-        "mathutils.interpolate",
-        "mathutils.noise",
-        # misc
-        "freestyle", "bgl", "blf",
-        "gpu", "gpu.offscreen",
-        "aud", "bpy_extras",
-        "idprop.types",
-        # bmesh, submodules are in own page
-        "bmesh",
+        # submodules are added in parent page
+        "mathutils", "freestyle", "bgl", "blf", "gpu",
+        "aud", "bpy_extras", "idprop.types", "bmesh",
     )
 
     for mod in standalone_modules:

doc/python_api/static/blender_logo.svg

@@ -0,0 +1,116 @@
+<?xml version="1.0" encoding="UTF-8" standalone="no"?>
+<svg
+   xmlns:dc="http://purl.org/dc/elements/1.1/"
+   xmlns:cc="http://creativecommons.org/ns#"
+   xmlns:rdf="http://www.w3.org/1999/02/22-rdf-syntax-ns#"
+   xmlns:svg="http://www.w3.org/2000/svg"
+   xmlns="http://www.w3.org/2000/svg"
+   id="svg8"
+   version="1.1"
+   viewBox="0 0 55.032989 15.935012"
+   height="60.226818"
+   width="207.9987">
+  <defs
+     id="defs2">
+    <clipPath
+       clipPathUnits="userSpaceOnUse"
+       id="clipPath3020">
+      <path
+         style="stroke-width:1.06666672"
+         d="M 0,0 H 211.2 V 61.866667 H 0 Z"
+         id="path3022" />
+    </clipPath>
+    <clipPath
+       clipPathUnits="userSpaceOnUse"
+       id="clipPath3020-8">
+      <path
+         style="stroke-width:1.06666672"
+         d="M 0,0 H 211.2 V 61.866667 H 0 Z"
+         id="path3022-1" />
+    </clipPath>
+  </defs>
+  <metadata
+     id="metadata5">
+    <rdf:RDF>
+      <cc:Work
+         rdf:about="">
+        <dc:format>image/svg+xml</dc:format>
+        <dc:type
+           rdf:resource="http://purl.org/dc/dcmitype/StillImage" />
+        <dc:creator>
+          <cc:Agent>
+            <dc:title>Blender Logo</dc:title>
+          </cc:Agent>
+        </dc:creator>
+        <dc:source>https://www.blender.org/about/logo/</dc:source>
+        <cc:license
+           rdf:resource="(c) Blender Foundation" />
+      </cc:Work>
+    </rdf:RDF>
+  </metadata>
+  <g
+     transform="translate(42.023693,-77.734934)"
+     id="layer1">
+    <g
+       transform="matrix(0.26259939,0,0,-0.26259939,-42.237694,93.888967)"
+       id="g3012"
+       style="stroke-width:7.2904439">
+      <g
+         id="g3014"
+         style="stroke-width:7.2904439" />
+      <g
+         id="g3016"
+         style="stroke-width:7.2904439">
+        <g
+           clip-path="url(#clipPath3020-8)"
+           id="g3018"
+           style="stroke-width:7.2904439">
+          <path
+             id="path3024"
+             style="fill:#ffffff;fill-opacity:1;fill-rule:nonzero;stroke:none;stroke-width:7.77647352"
+             d="m 192.54827,44.510933 c 0,0 -125.158403,0 -128.688003,0 -0.07893,0.06187 -0.1504,0.1344 -0.2272,0.193067 -0.01813,0.0192 -18.9312,14.548267 -19.5008,14.986667 -0.032,0.0256 -0.06507,0.04907 -0.06507,0.04907 -3.029333,2.321067 -7.531733,2.369067 -10.721067,0.133334 -1.9968,-1.396267 -3.194666,-3.508267 -3.285333,-5.792 -0.0021,-0.09173 -0.0064,-0.1824 -0.0064,-0.2752 0,-1.185067 0.314667,-2.3296 0.8832,-3.357867 -5.857067,-0.0053 -11.746133,-0.0128 -11.746133,-0.0128 -4.3936,-0.0021 -8.3648,-2.965333 -9.2725337,-6.8928 C 9.7898667,42.9792 9.7248,42.411733 9.7248,41.8496 c 0,-1.643733 0.546133,-3.236267 1.5808,-4.542933 1.162667,-1.463467 2.842667,-2.448 4.7584,-2.832 C 10.530133,30.2272 5.0016,25.984 4.9973333,25.979733 4.9578667,25.949867 4.9248,25.924267 4.8970667,25.905067 c -2.2176,-1.703467 -3.68,-4.1728 -4.0106667,-6.770134 -0.0490667,-0.381866 -0.0714667,-0.7584 -0.0714667,-1.133866 0,-1.870934 0.5962667,-3.650134 1.7301334,-5.102934 1.3781333,-1.764266 3.4144,-2.884266 5.7322666,-3.147733 2.6549337,-0.3072 5.4495997,0.542933 7.6607997,2.330667 0.01493,0.01173 2.753067,2.256 5.639467,4.6176 1.060267,-2.555734 2.545067,-4.926934 4.465067,-7.0453337 2.4704,-2.7306666 5.473066,-4.8864 8.919466,-6.4128 C 38.5856,1.6352 42.466133,0.82453333 46.501333,0.83413333 50.5376,0.8416 54.417067,1.6629333 58.0352,3.2768 c 3.453867,1.5466667 6.449067,3.7109333 8.910933,6.4352 0.360534,0.4032 0.693334,0.8288 1.0272,1.2544 4.885334,0 124.574937,0 124.574937,0 9.84,0 17.8368,7.5232 17.8368,16.7712 0,9.248 -7.9968,16.773333 -17.8368,16.773333" />
+          <path
+             id="path3026"
+             style="fill:#0d528a;fill-opacity:1;fill-rule:nonzero;stroke:none;stroke-width:7.77647352"
+             d="m 38.538667,28.497067 c 0.113066,2.016 1.101866,3.793066 2.590933,5.0528 1.4624,1.237333 3.428267,1.9936 5.575467,1.9936 2.144,0 4.110933,-0.756267 5.572266,-1.9936 1.490134,-1.259734 2.4768,-3.0368 2.590934,-5.050667 0.113066,-2.0736 -0.718934,-3.997867 -2.181334,-5.425067 C 51.1968,21.6224 49.0784,20.7104 46.705067,20.7104 c -2.3744,0 -4.497067,0.912 -5.9872,2.363733 -1.461334,1.4272 -2.292267,3.351467 -2.1792,5.422934" />
+          <path
+             id="path3028"
+             style="fill:#f5792a;fill-opacity:1;fill-rule:nonzero;stroke:none;stroke-width:7.77647352"
+             d="m 25.536,24.421333 c 0.01387,-0.7904 0.2656,-2.3232 0.6432,-3.521066 C 26.9728,18.3648 28.318933,16.0192 30.193067,13.950933 32.1152,11.8272 34.482133,10.119467 37.216,8.9088 c 2.872533,-1.2736 5.986133,-1.9210667 9.220267,-1.9157333 3.229866,0.00427 6.3424,0.6656 9.216,1.9477333 2.733866,1.2224 5.098666,2.9376 7.018666,5.063467 1.870934,2.074666 3.216,4.424533 4.010667,6.96 0.401067,1.282133 0.6528,2.581333 0.754133,3.886933 0.100267,1.285333 0.0576,2.571733 -0.1248,3.858133 -0.356266,2.507734 -1.223466,4.858667 -2.557866,7.002667 -1.2224,1.970133 -2.7968,3.696 -4.6688,5.147733 l 0.0053,0.0021 -18.8928,14.506666 c -0.016,0.0128 -0.02987,0.02667 -0.048,0.03733 -1.240533,0.952533 -3.3248,0.948266 -4.686933,-0.0053 -1.3792,-0.9632 -1.536,-2.557866 -0.3104,-3.5648 l -0.0043,-0.0043 7.8784,-6.407467 -24.016,-0.02667 c -0.01173,0 -0.0224,0 -0.032,0 -1.985067,-0.0011 -3.893333,-1.303466 -4.269867,-2.9504 -0.389333,-1.6768 0.958934,-3.067733 3.022934,-3.074133 l -0.0021,-0.0075 L 30.900267,39.3888 9.1786667,22.715733 c -0.026667,-0.02133 -0.0576,-0.0416 -0.0832,-0.06293 -2.0490667,-1.568 -2.7104,-4.178133 -1.4197334,-5.828267 1.3088,-1.68 4.0949337,-1.6832 6.1653337,-0.0096 L 25.696,26.516267 c 0,0 -0.173867,-1.3088 -0.16,-2.094934 z M 56,20.034133 c -2.443733,-2.488533 -5.8624,-3.899733 -9.563733,-3.9072 -3.7056,-0.0064 -7.124267,1.393067 -9.568,3.877334 -1.1936,1.210666 -2.0704,2.602666 -2.6112,4.087466 -0.529067,1.457067 -0.736,3.0048 -0.599467,4.5664 0.130133,1.527467 0.583467,2.9824 1.309867,4.3008 0.712533,1.293867 1.6928,2.465067 2.9056,3.454934 2.373333,1.934933 5.396266,2.981333 8.558933,2.9856 3.1648,0.0053 6.1856,-1.0336 8.561067,-2.961067 1.2096,-0.9856 2.190933,-2.151467 2.903466,-3.445333 0.728534,-1.316267 1.179734,-2.765867 1.314134,-4.2976 C 59.344,27.136 59.138133,25.5904 58.609067,24.1312 58.0672,22.6432 57.1936,21.2512 56,20.034133" />
+          <path
+             id="path3030"
+             style="fill:#0d528a;fill-opacity:1;fill-rule:nonzero;stroke:none;stroke-width:7.77647352"
+             d="m 167.96587,28.529067 c 0.1824,2.0352 1.75466,3.325866 4.15786,3.325866 2.4064,0 3.97974,-1.290666 4.16214,-3.325866 z M 176.0384,23.504 c -0.61547,-1.508267 -2.06507,-2.401067 -4.03627,-2.401067 -2.46613,0 -4.12693,1.540267 -4.1952,3.9744 h 14.18347 c 0,0.251734 0,0.462934 0,0.715734 0,6.094933 -3.57653,9.518933 -9.98827,9.518933 -6.22506,0 -9.98826,-3.457067 -9.98826,-8.878933 0,-5.454934 3.82293,-8.9056 9.98826,-8.9056 3.70134,0 6.5696,1.2672 8.33814,3.4976 L 176.0384,23.504" />
+          <path
+             id="path3032"
+             style="fill:#0d528a;fill-opacity:1;fill-rule:nonzero;stroke:none;stroke-width:7.77647352"
+             d="m 105.0336,28.519467 c 0.18667,2.036266 1.76,3.326933 4.16533,3.326933 2.40214,0 3.97547,-1.290667 4.15894,-3.326933 z m 8.07787,-5.025067 c -0.61547,-1.512533 -2.06614,-2.404267 -4.03947,-2.404267 -2.464,0 -4.12907,1.544534 -4.19413,3.9776 h 14.18026 c 0,0.2496 0,0.4608 0,0.712534 0,6.098133 -3.57546,9.524266 -9.98613,9.524266 -6.22933,0 -9.986133,-3.458133 -9.986133,-8.88 0,-5.456 3.821863,-8.906666 9.986133,-8.906666 3.6992,0 6.5696,1.262933 8.34133,3.495466 l -4.30186,2.481067" />
+          <path
+             id="path3034"
+             style="fill:#0d528a;fill-opacity:1;fill-rule:nonzero;stroke:none;stroke-width:7.77647352"
+             d="m 92.701867,38.997333 h 5.114666 v -20.9248 h -5.114666 z" />
+          <path
+             id="path3036"
+             style="fill:#0d528a;fill-opacity:1;fill-rule:nonzero;stroke:none;stroke-width:7.77647352"
+             d="m 120.33067,33.438933 h 5.14453 v -1.245866 c 1.8208,1.9424 4.00747,2.9248 6.47253,2.9248 2.84054,0 4.992,-0.9824 6.1024,-2.653867 0.92587,-1.381333 0.98774,-3.0496 0.98774,-5.239467 v -9.152 h -5.15094 v 8.040534 c 0,3.336533 -0.67093,4.877866 -3.6,4.877866 -2.96106,0 -4.81173,-1.764266 -4.81173,-4.724266 v -8.194134 h -5.14453 v 15.3664" />
+          <path
+             id="path3038"
+             style="fill:#0d528a;fill-opacity:1;fill-rule:nonzero;stroke:none;stroke-width:7.77647352"
+             d="m 155.60107,26.173867 c 0,-2.864 -1.9136,-4.741334 -4.8736,-4.741334 -2.96427,0 -4.87787,1.813334 -4.87787,4.709334 0,2.9376 1.88907,4.750933 4.87787,4.750933 2.96,0 4.8736,-1.8464 4.8736,-4.718933 z m 0,6.568533 c -1.30027,1.393067 -3.1808,2.157867 -5.79734,2.157867 -5.64266,0 -9.49546,-3.479467 -9.49546,-8.6944 0,-5.112534 3.82613,-8.688 9.40266,-8.688 2.5568,0 4.4352,0.645333 5.89014,2.096 v -1.541334 h 5.14666 v 22.2528 l -5.14666,-1.328 V 32.7424" />
+          <path
+             id="path3040"
+             style="fill:#0d528a;fill-opacity:1;fill-rule:nonzero;stroke:none;stroke-width:7.77647352"
+             d="m 81.010133,30.8928 c 2.990934,0 4.871467,-1.813333 4.871467,-4.750933 0,-2.896 -1.9104,-4.709334 -4.871467,-4.709334 -2.958933,0 -4.869333,1.877334 -4.869333,4.741334 0,2.872533 1.9104,4.718933 4.869333,4.718933 z m -4.869333,8.104533 -5.1488,1.328 v -22.2528 h 5.1488 v 1.541334 c 1.448533,-1.450667 3.329067,-2.096 5.886933,-2.096 5.579734,0 9.4016,3.575466 9.4016,8.688 0,5.214933 -3.853866,8.6944 -9.493333,8.6944 -2.621867,0 -4.5024,-0.7648 -5.7952,-2.157867 v 6.254933" />
+          <path
+             id="path3042"
+             style="fill:#0d528a;fill-opacity:1;fill-rule:nonzero;stroke:none;stroke-width:7.77647352"
+             d="m 183.2608,18.072533 v 15.3664 h 5.14667 v -0.9504 c 1.54026,1.857067 3.1712,2.7808 5.0272,2.7808 0.368,0 0.82986,-0.05973 1.44533,-0.1216 V 30.768 c -0.496,0.064 -1.04747,0.064 -1.6352,0.064 -2.992,0 -4.83733,-1.972267 -4.83733,-5.329067 v -7.4304 h -5.14667" />
+          <path
+             id="path3044"
+             style="fill:#0d528a;fill-opacity:1;fill-rule:nonzero;stroke:none;stroke-width:7.77647352"
+             d="m 201.55307,35.252267 h -0.39467 l 0.0139,1.409066 0.0203,0.466134 -0.1056,-0.384 -0.46826,-1.4912 h -0.36054 l -0.45866,1.4912 -0.1152,0.376533 0.0277,-0.458667 0.0139,-1.409066 h -0.38614 v 2.282666 h 0.54187 l 0.57067,-1.8016 0.5568,1.8016 h 0.544 z m -3.12747,0 h -0.39787 V 37.1968 h -0.7328 v 0.338133 h 1.85067 V 37.1968 h -0.72 v -1.944533" />
+        </g>
+      </g>
+    </g>
+  </g>
+</svg>

extern/carve/include/carve/win32.h

@@ -8,11 +8,9 @@
 #include <string.h>
 #include <stdlib.h>
 
-#if !defined(__MINGW32__)
 inline int strcasecmp(const char *a, const char *b) {
   return _stricmp(a,b);
 }
-#endif
 
 inline void srandom(unsigned long input) {
   srand(input);
@@ -34,7 +32,7 @@ typedef unsigned long uintptr_t;
 
 #  if _MSC_VER < 1600
 // stdint.h is not available before VS2010
-#if defined(_WIN32) && !defined(__MINGW32__)
+#if defined(_WIN32)
 /* The __intXX are built-in types of the visual complier! So we don't
    need to include anything else here.
    This typedefs should be in sync with types from MEM_sys_types.h */

extern/carve/patches/mingw.patch

@@ -1,15 +0,0 @@
-diff -r 525472fb477a include/carve/win32.h
---- a/include/carve/win32.h	Sun Jan 15 23:07:40 2012 -0500
-+++ b/include/carve/win32.h	Wed Jan 18 00:40:10 2012 +0600
-@@ -8,9 +8,11 @@
- #include <string.h>
- #include <stdlib.h>
- 
-+#if !defined(__MINGW32__)
- inline int strcasecmp(const char *a, const char *b) {
-   return _stricmp(a,b);
- }
-+#endif
- 
- inline void srandom(unsigned long input) {
-   srand(input);

extern/carve/patches/mingw_w64.patch

@@ -1,13 +0,0 @@
-Index: bundle.sh
-===================================================================
---- bundle.sh	(revision 45912)
-+++ bundle.sh	(working copy)
-@@ -114,7 +114,7 @@
- if env['WITH_BF_BOOST']:
-     if env['OURPLATFORM'] not in ('win32-vc', 'win64-vc'):
-         # Boost is setting as preferred collections library in the Carve code when using MSVC compiler
--        if env['OURPLATFORM'] != 'win32-mingw':
-+        if env['OURPLATFORM'] not in ('win32-mingw', 'win64-mingw'):
-             defs.append('HAVE_BOOST_UNORDERED_COLLECTIONS')
- 
-     defs.append('CARVE_SYSTEM_BOOST')

extern/carve/patches/series

@@ -1,7 +1,6 @@
 includes.patch
 win32.patch
 mesh_iterator.patch
-mingw.patch
 gcc46.patch
 clang_is_heap_fix.patch
 strict_flags.patch

extern/carve/patches/win32.patch

@@ -8,7 +8,7 @@ diff -r e82d852e4fb0 include/carve/win32.h
 -typedef char int8_t;
 -typedef short int16_t;
 -typedef long int32_t;
-+#if defined(_WIN32) && !defined(__MINGW32__)
++#if defined(_WIN32)
 +/* The __intXX are built-in types of the visual complier! So we don't
 +   need to include anything else here.
 +   This typedefs should be in sync with types from MEM_sys_types.h */

extern/curve_fit_nd/intern/curve_fit_cubic_refit.c

@@ -137,7 +137,7 @@ struct Knot {
 
 	/* Initially point to contiguous memory, however we may re-assign */
 	double *tan[2];
-} Knot;
+};
 
 
 struct KnotRemoveState {
@@ -1102,7 +1102,7 @@ int curve_fit_cubic_to_points_refit_db(
         uint   **r_corner_index_array, uint *r_corner_index_len)
 {
 	const uint knots_len = points_len;
-	struct Knot *knots = malloc(sizeof(Knot) * knots_len);
+	struct Knot *knots = malloc(sizeof(struct Knot) * knots_len);
 
 #ifndef USE_CORNER_DETECT
 	(void)r_corner_index_array;

intern/atomic/intern/atomic_ops_utils.h

@@ -63,12 +63,7 @@
 #if defined(_MSC_VER)
 #  define ATOMIC_INLINE static __forceinline
 #else
-#  if (defined(__APPLE__) && defined(__ppc__))
-/* static inline __attribute__ here breaks osx ppc gcc42 build */
-#    define ATOMIC_INLINE static __attribute__((always_inline))
-#  else
-#    define ATOMIC_INLINE static inline __attribute__((always_inline))
-#  endif
+#  define ATOMIC_INLINE static inline __attribute__((always_inline))
 #endif
 
 #ifndef LIKELY

intern/audaspace/CMakeLists.txt

@@ -19,7 +19,7 @@
 #
 # ***** END LGPL LICENSE BLOCK *****
 
-remove_extra_strict_flags()
+remove_strict_flags()
 
 if(CMAKE_COMPILER_IS_GNUCC)
 	remove_cc_flag("-Wunused-macros")

intern/audaspace/intern/AUD_SoftwareDevice.cpp

@@ -89,7 +89,7 @@ bool AUD_SoftwareDevice::AUD_SoftwareHandle::pause(bool keep)
 }
 
 AUD_SoftwareDevice::AUD_SoftwareHandle::AUD_SoftwareHandle(AUD_SoftwareDevice* device, boost::shared_ptr<AUD_IReader> reader, boost::shared_ptr<AUD_PitchReader> pitch, boost::shared_ptr<AUD_ResampleReader> resampler, boost::shared_ptr<AUD_ChannelMapperReader> mapper, bool keep) :
-	m_reader(reader), m_pitch(pitch), m_resampler(resampler), m_mapper(mapper), m_keep(keep), m_user_pitch(1.0f), m_user_volume(1.0f), m_user_pan(0.0f), m_volume(1.0f), m_old_volume(1.0f), m_loopcount(0),
+	m_reader(reader), m_pitch(pitch), m_resampler(resampler), m_mapper(mapper), m_keep(keep), m_user_pitch(1.0f), m_user_volume(1.0f), m_user_pan(0.0f), m_volume(0.0f), m_old_volume(0.0f), m_loopcount(0),
 	m_relative(true), m_volume_max(1.0f), m_volume_min(0), m_distance_max(std::numeric_limits<float>::max()),
 	m_distance_reference(1.0f), m_attenuation(1.0f), m_cone_angle_outer(M_PI), m_cone_angle_inner(M_PI), m_cone_volume_outer(0),
 	m_flags(AUD_RENDER_CONE), m_stop(NULL), m_stop_data(NULL), m_status(AUD_STATUS_PLAYING), m_device(device)

intern/cycles/blender/addon/engine.py

@@ -239,14 +239,15 @@ def register_passes(engine, scene, srl):
     if crl.pass_debug_bvh_intersections:       engine.register_pass(scene, srl, "Debug BVH Intersections",       1, "X", 'VALUE')
     if crl.pass_debug_ray_bounces:             engine.register_pass(scene, srl, "Debug Ray Bounces",             1, "X", 'VALUE')
 
-    if crl.use_denoising and crl.denoising_store_passes:
-        engine.register_pass(scene, srl, "Denoising Normal",          3, "XYZ", 'VECTOR');
-        engine.register_pass(scene, srl, "Denoising Normal Variance", 3, "XYZ", 'VECTOR');
-        engine.register_pass(scene, srl, "Denoising Albedo",          3, "RGB", 'COLOR');
-        engine.register_pass(scene, srl, "Denoising Albedo Variance", 3, "RGB", 'COLOR');
-        engine.register_pass(scene, srl, "Denoising Depth",           1, "Z",   'VALUE');
-        engine.register_pass(scene, srl, "Denoising Depth Variance",  1, "Z",   'VALUE');
-        engine.register_pass(scene, srl, "Denoising Shadow A",        3, "XYV", 'VECTOR');
-        engine.register_pass(scene, srl, "Denoising Shadow B",        3, "XYV", 'VECTOR');
-        engine.register_pass(scene, srl, "Denoising Image",           3, "RGB", 'COLOR');
-        engine.register_pass(scene, srl, "Denoising Image Variance",  3, "RGB", 'COLOR');
+    cscene = scene.cycles
+    if crl.use_denoising and crl.denoising_store_passes and not cscene.use_progressive_refine:
+        engine.register_pass(scene, srl, "Denoising Normal",          3, "XYZ", 'VECTOR')
+        engine.register_pass(scene, srl, "Denoising Normal Variance", 3, "XYZ", 'VECTOR')
+        engine.register_pass(scene, srl, "Denoising Albedo",          3, "RGB", 'COLOR')
+        engine.register_pass(scene, srl, "Denoising Albedo Variance", 3, "RGB", 'COLOR')
+        engine.register_pass(scene, srl, "Denoising Depth",           1, "Z",   'VALUE')
+        engine.register_pass(scene, srl, "Denoising Depth Variance",  1, "Z",   'VALUE')
+        engine.register_pass(scene, srl, "Denoising Shadow A",        3, "XYV", 'VECTOR')
+        engine.register_pass(scene, srl, "Denoising Shadow B",        3, "XYV", 'VECTOR')
+        engine.register_pass(scene, srl, "Denoising Image",           3, "RGB", 'COLOR')
+        engine.register_pass(scene, srl, "Denoising Image Variance",  3, "RGB", 'COLOR')

intern/cycles/blender/addon/properties.py

@@ -695,10 +695,17 @@ class CyclesRenderSettings(bpy.types.PropertyGroup):
             update=devices_update_callback
             )
 
-        cls.debug_opencl_kernel_single_program = BoolProperty(name="Single Program", default=True, update=devices_update_callback);
+        cls.debug_opencl_kernel_single_program = BoolProperty(
+            name="Single Program",
+            default=True,
+            update=devices_update_callback,
+            )
 
         cls.debug_use_opencl_debug = BoolProperty(name="Debug OpenCL", default=False)
 
+        cls.debug_opencl_mem_limit = IntProperty(name="Memory limit", default=0,
+            description="Artificial limit on OpenCL memory usage in MB (0 to disable limit)")
+
     @classmethod
     def unregister(cls):
         del bpy.types.Scene.cycles
@@ -1166,6 +1173,12 @@ class CyclesCurveRenderSettings(bpy.types.PropertyGroup):
     def unregister(cls):
         del bpy.types.Scene.cycles_curves
 
+def update_render_passes(self, context):
+    scene = context.scene
+    rd = scene.render
+    rl = rd.layers.active
+    rl.update_render_passes()
+
 class CyclesRenderLayerSettings(bpy.types.PropertyGroup):
     @classmethod
     def register(cls):
@@ -1178,27 +1191,32 @@ class CyclesRenderLayerSettings(bpy.types.PropertyGroup):
                 name="Debug BVH Traversed Nodes",
                 description="Store Debug BVH Traversed Nodes pass",
                 default=False,
+                update=update_render_passes,
                 )
         cls.pass_debug_bvh_traversed_instances = BoolProperty(
                 name="Debug BVH Traversed Instances",
                 description="Store Debug BVH Traversed Instances pass",
                 default=False,
+                update=update_render_passes,
                 )
         cls.pass_debug_bvh_intersections = BoolProperty(
                 name="Debug BVH Intersections",
                 description="Store Debug BVH Intersections",
                 default=False,
+                update=update_render_passes,
                 )
         cls.pass_debug_ray_bounces = BoolProperty(
                 name="Debug Ray Bounces",
                 description="Store Debug Ray Bounces pass",
                 default=False,
+                update=update_render_passes,
                 )
 
         cls.use_denoising = BoolProperty(
                 name="Use Denoising",
                 description="Denoise the rendered image",
                 default=False,
+                update=update_render_passes,
                 )
         cls.denoising_diffuse_direct = BoolProperty(
                 name="Diffuse Direct",
@@ -1255,18 +1273,19 @@ class CyclesRenderLayerSettings(bpy.types.PropertyGroup):
         cls.denoising_radius = IntProperty(
                 name="Denoising Radius",
                 description="Size of the image area that's used to denoise a pixel (higher values are smoother, but might lose detail and are slower)",
-                min=1, max=50,
+                min=1, max=25,
                 default=8,
         )
         cls.denoising_relative_pca = BoolProperty(
                 name="Relative filter",
-                description="When removing that don't carry information, use a relative threshold instead of an absolute one (can help to reduce artifacts, but might cause detail loss around edges)",
+                description="When removing pixels that don't carry information, use a relative threshold instead of an absolute one (can help to reduce artifacts, but might cause detail loss around edges)",
                 default=False,
         )
         cls.denoising_store_passes = BoolProperty(
                 name="Store denoising passes",
                 description="Store the denoising feature passes and the noisy image",
                 default=False,
+                update=update_render_passes,
         )
 
     @classmethod

intern/cycles/blender/addon/ui.py

@@ -531,17 +531,17 @@ class CyclesRender_PT_layer_passes(CyclesButtonsPanel, Panel):
         col.prop(rl, "use_pass_environment")
 
         if context.scene.cycles.feature_set == 'EXPERIMENTAL':
-           col.separator()
-           sub = col.column()
-           sub.active = crl.use_denoising
-           sub.prop(crl, "denoising_store_passes", text="Denoising")
+            col.separator()
+            sub = col.column()
+            sub.active = crl.use_denoising
+            sub.prop(crl, "denoising_store_passes", text="Denoising")
 
         if _cycles.with_cycles_debug:
-          col = layout.column()
-          col.prop(crl, "pass_debug_bvh_traversed_nodes")
-          col.prop(crl, "pass_debug_bvh_traversed_instances")
-          col.prop(crl, "pass_debug_bvh_intersections")
-          col.prop(crl, "pass_debug_ray_bounces")
+            col = layout.column()
+            col.prop(crl, "pass_debug_bvh_traversed_nodes")
+            col.prop(crl, "pass_debug_bvh_traversed_instances")
+            col.prop(crl, "pass_debug_bvh_intersections")
+            col.prop(crl, "pass_debug_ray_bounces")
 
 
 class CyclesRender_PT_views(CyclesButtonsPanel, Panel):
@@ -596,16 +596,23 @@ class CyclesRender_PT_denoising(CyclesButtonsPanel, Panel):
         rd = context.scene.render
         rl = rd.layers.active
         crl = rl.cycles
-        self.layout.prop(crl, "use_denoising", text="")
+        cscene = context.scene.cycles
+        layout = self.layout
+
+        layout.active = not cscene.use_progressive_refine
+        layout.prop(crl, "use_denoising", text="")
 
     def draw(self, context):
         layout = self.layout
 
         scene = context.scene
+        cscene = scene.cycles
         rd = scene.render
         rl = rd.layers.active
         crl = rl.cycles
 
+        layout.active = crl.use_denoising and not cscene.use_progressive_refine
+
         split = layout.split()
 
         col = split.column()
@@ -1601,6 +1608,7 @@ class CyclesRender_PT_debug(CyclesButtonsPanel, Panel):
         col.prop(cscene, "debug_opencl_device_type", text="Device")
         col.prop(cscene, "debug_opencl_kernel_single_program", text="Single Program")
         col.prop(cscene, "debug_use_opencl_debug", text="Debug")
+        col.prop(cscene, "debug_opencl_mem_limit")
 
 
 class CyclesParticle_PT_CurveSettings(CyclesButtonsPanel, Panel):
@@ -1703,7 +1711,7 @@ def draw_device(self, context):
 
         layout.prop(cscene, "feature_set")
 
-        split = layout.split(percentage=1/3)
+        split = layout.split(percentage=1 / 3)
         split.label("Device:")
         row = split.row()
         row.active = show_device_active(context)

intern/cycles/blender/blender_camera.cpp

@@ -544,7 +544,11 @@ void BlenderSync::sync_camera_motion(BL::RenderSettings& b_render,
 
 	if(tfm != cam->matrix) {
 		VLOG(1) << "Camera " << b_ob.name() << " motion detected.";
-		if(motion_time == -1.0f) {
+		if(motion_time == 0.0f) {
+			/* When motion blur is not centered in frame, cam->matrix gets reset. */
+			cam->matrix = tfm;
+		}
+		else if(motion_time == -1.0f) {
 			cam->motion.pre = tfm;
 			cam->use_motion = true;
 		}
@@ -573,7 +577,10 @@ void BlenderSync::sync_camera_motion(BL::RenderSettings& b_render,
 		float fov = 2.0f * atanf((0.5f * sensor_size) / bcam.lens / aspectratio);
 		if(fov != cam->fov) {
 			VLOG(1) << "Camera " << b_ob.name() << " FOV change detected.";
-			if(motion_time == -1.0f) {
+			if(motion_time == 0.0f) {
+				cam->fov = fov;
+			}
+			else if(motion_time == -1.0f) {
 				cam->fov_pre = fov;
 				cam->use_perspective_motion = true;
 			}

intern/cycles/blender/blender_mesh.cpp

@@ -14,7 +14,6 @@
  * limitations under the License.
  */
 
- 
 #include "render/mesh.h"
 #include "render/object.h"
 #include "render/scene.h"
@@ -51,8 +50,7 @@ enum {
  * Two triangles has vertex indices in the original Blender-side face.
  * If face is already a quad tri_b will not be initialized.
  */
-inline void face_split_tri_indices(const int num_verts,
-                                   const int face_flag,
+inline void face_split_tri_indices(const int face_flag,
                                    int tri_a[3],
                                    int tri_b[3])
 {
@@ -60,21 +58,19 @@ inline void face_split_tri_indices(const int num_verts,
 		tri_a[0] = 0;
 		tri_a[1] = 1;
 		tri_a[2] = 3;
-		if(num_verts == 4) {
-			tri_b[0] = 2;
-			tri_b[1] = 3;
-			tri_b[2] = 1;
-		}
+
+		tri_b[0] = 2;
+		tri_b[1] = 3;
+		tri_b[2] = 1;
 	}
 	else /*if(face_flag & FACE_FLAG_DIVIDE_13)*/ {
 		tri_a[0] = 0;
 		tri_a[1] = 1;
 		tri_a[2] = 2;
-		if(num_verts == 4) {
-			tri_b[0] = 0;
-			tri_b[1] = 2;
-			tri_b[2] = 3;
-		}
+
+		tri_b[0] = 0;
+		tri_b[1] = 2;
+		tri_b[2] = 3;
 	}
 }
 
@@ -251,7 +247,7 @@ static void mikk_compute_tangents(BL::Mesh& b_mesh,
 
 	for(int i = 0; i < nverts.size(); i++) {
 		int tri_a[3], tri_b[3];
-		face_split_tri_indices(nverts[i], face_flags[i], tri_a, tri_b);
+		face_split_tri_indices(face_flags[i], tri_a, tri_b);
 
 		tangent[0] = float4_to_float3(userdata.tangent[i*4 + tri_a[0]]);
 		tangent[1] = float4_to_float3(userdata.tangent[i*4 + tri_a[1]]);
@@ -293,7 +289,7 @@ static void create_mesh_volume_attribute(BL::Object& b_ob,
 
 	if(!b_domain)
 		return;
-	
+
 	Attribute *attr = mesh->attributes.add(std);
 	VoxelAttribute *volume_data = attr->data_voxel();
 	bool is_float, is_linear;
@@ -377,7 +373,7 @@ static void attr_create_vertex_color(Scene *scene,
 
 			for(l->data.begin(c); c != l->data.end(); ++c, ++i) {
 				int tri_a[3], tri_b[3];
-				face_split_tri_indices(nverts[i], face_flags[i], tri_a, tri_b);
+				face_split_tri_indices(face_flags[i], tri_a, tri_b);
 
 				uchar4 colors[4];
 				colors[0] = color_float_to_byte(color_srgb_to_scene_linear_v3(get_float3(c->color1())));
@@ -470,7 +466,7 @@ static void attr_create_uv_map(Scene *scene,
 
 				for(l->data.begin(t); t != l->data.end(); ++t, ++i) {
 					int tri_a[3], tri_b[3];
-					face_split_tri_indices(nverts[i], face_flags[i], tri_a, tri_b);
+					face_split_tri_indices(face_flags[i], tri_a, tri_b);
 
 					float3 uvs[4];
 					uvs[0] = get_float3(t->uv1());
@@ -982,7 +978,7 @@ Mesh *BlenderSync::sync_mesh(BL::Object& b_ob,
 		else
 			used_shaders.push_back(scene->default_surface);
 	}
-	
+
 	/* test if we need to sync */
 	int requested_geometry_flags = Mesh::GEOMETRY_NONE;
 	if(render_layer.use_surfaces) {
@@ -1017,12 +1013,12 @@ Mesh *BlenderSync::sync_mesh(BL::Object& b_ob,
 	/* ensure we only sync instanced meshes once */
 	if(mesh_synced.find(mesh) != mesh_synced.end())
 		return mesh;
-	
+
 	mesh_synced.insert(mesh);
 
 	/* create derived mesh */
 	array<int> oldtriangle = mesh->triangles;
-	
+
 	/* compares curve_keys rather than strands in order to handle quick hair
 	 * adjustments in dynamic BVH - other methods could probably do this better*/
 	array<float3> oldcurve_keys = mesh->curve_keys;
@@ -1111,7 +1107,7 @@ Mesh *BlenderSync::sync_mesh(BL::Object& b_ob,
 		if(memcmp(&oldcurve_radius[0], &mesh->curve_radius[0], sizeof(float)*oldcurve_radius.size()) != 0)
 			rebuild = true;
 	}
-	
+
 	mesh->tag_update(scene, rebuild);
 
 	return mesh;
@@ -1140,7 +1136,7 @@ void BlenderSync::sync_mesh_motion(BL::Object& b_ob,
 	if(scene->need_motion() == Scene::MOTION_BLUR) {
 		if(!mesh->use_motion_blur)
 			return;
-		
+
 		/* see if this mesh needs motion data at this time */
 		vector<float> object_times = object->motion_times();
 		bool found = false;
@@ -1172,7 +1168,7 @@ void BlenderSync::sync_mesh_motion(BL::Object& b_ob,
 
 	if(!numverts && !numkeys)
 		return;
-	
+
 	/* skip objects without deforming modifiers. this is not totally reliable,
 	 * would need a more extensive check to see which objects are animated */
 	BL::Mesh b_mesh(PointerRNA_NULL);

intern/cycles/blender/blender_python.cpp

@@ -106,6 +106,7 @@ bool debug_flags_sync_from_scene(BL::Scene b_scene)
 	}
 	/* Synchronize other OpenCL flags. */
 	flags.opencl.debug = get_boolean(cscene, "debug_use_opencl_debug");
+	flags.opencl.mem_limit = ((size_t)get_int(cscene, "debug_opencl_mem_limit"))*1024*1024;
 	flags.opencl.single_program = get_boolean(cscene, "debug_opencl_kernel_single_program");
 	return flags.opencl.device_type != opencl_device_type ||
 	       flags.opencl.kernel_type != opencl_kernel_type;

intern/cycles/blender/blender_session.cpp

@@ -129,9 +129,9 @@ void BlenderSession::create_session()
 	scene = new Scene(scene_params, session_params.device);
 
 	/* setup callbacks for builtin image support */
-	scene->image_manager->builtin_image_info_cb = function_bind(&BlenderSession::builtin_image_info, this, _1, _2, _3, _4, _5, _6, _7);
-	scene->image_manager->builtin_image_pixels_cb = function_bind(&BlenderSession::builtin_image_pixels, this, _1, _2, _3, _4);
-	scene->image_manager->builtin_image_float_pixels_cb = function_bind(&BlenderSession::builtin_image_float_pixels, this, _1, _2, _3, _4);
+	scene->image_manager->builtin_image_info_cb = function_bind(&BlenderSession::builtin_image_info, this, _1, _2, _3, _4, _5, _6, _7, _8);
+	scene->image_manager->builtin_image_pixels_cb = function_bind(&BlenderSession::builtin_image_pixels, this, _1, _2, _3, _4, _5);
+	scene->image_manager->builtin_image_float_pixels_cb = function_bind(&BlenderSession::builtin_image_float_pixels, this, _1, _2, _3, _4, _5);
 
 	/* create session */
 	session = new Session(session_params);
@@ -399,20 +399,14 @@ void BlenderSession::render()
 		BL::RenderLayer b_rlay = *b_single_rlay;
 
 		/* add passes */
-		array<Pass> passes;
-		if(session_params.device.advanced_shading) {
-			passes = sync->sync_render_passes(b_rlay, *b_layer_iter);
-		}
-		else {
-			Pass::add(PASS_COMBINED, passes);
-		}
-
+		array<Pass> passes = sync->sync_render_passes(b_rlay, *b_layer_iter, session_params);
 		buffer_params.passes = passes;
 
 		PointerRNA crl = RNA_pointer_get(&b_layer_iter->ptr, "cycles");
-		buffer_params.denoising_data_pass = get_boolean(crl, "use_denoising");
-		session->tile_manager.schedule_denoising = get_boolean(crl, "use_denoising");
-		session->params.use_denoising = get_boolean(crl, "use_denoising");
+		bool use_denoising = !session_params.progressive_refine && get_boolean(crl, "use_denoising");
+		buffer_params.denoising_data_pass = use_denoising;
+		session->tile_manager.schedule_denoising = use_denoising;
+		session->params.use_denoising = use_denoising;
 		scene->film->denoising_data_pass = buffer_params.denoising_data_pass;
 		scene->film->denoising_flags = 0;
 		if(!get_boolean(crl, "denoising_diffuse_direct"))        scene->film->denoising_flags |= DENOISING_CLEAN_DIFFUSE_DIR;
@@ -564,8 +558,6 @@ void BlenderSession::bake(BL::Object& b_object,
                           float result[])
 {
 	ShaderEvalType shader_type = get_shader_type(pass_type);
-	size_t object_index = OBJECT_NONE;
-	int tri_offset = 0;
 
 	/* Set baking flag in advance, so kernel loading can check if we need
 	 * any baking capabilities.
@@ -575,9 +567,6 @@ void BlenderSession::bake(BL::Object& b_object,
 	/* ensure kernels are loaded before we do any scene updates */
 	session->load_kernels();
 
-	if(session->progress.get_cancel())
-		return;
-
 	if(shader_type == SHADER_EVAL_UV) {
 		/* force UV to be available */
 		Pass::add(PASS_UV, scene->film->passes);
@@ -595,50 +584,61 @@ void BlenderSession::bake(BL::Object& b_object,
 	scene->film->tag_update(scene);
 	scene->integrator->tag_update(scene);
 
-	/* update scene */
-	BL::Object b_camera_override(b_engine.camera_override());
-	sync->sync_camera(b_render, b_camera_override, width, height, "");
-	sync->sync_data(b_render,
-	                b_v3d,
-	                b_camera_override,
-	                width, height,
-	                &python_thread_state,
-	                b_rlay_name.c_str());
-
-	/* get buffer parameters */
-	SessionParams session_params = BlenderSync::get_session_params(b_engine, b_userpref, b_scene, background);
-	BufferParams buffer_params = BlenderSync::get_buffer_params(b_render, b_v3d, b_rv3d, scene->camera, width, height);
-
-	scene->bake_manager->set_shader_limit((size_t)b_engine.tile_x(), (size_t)b_engine.tile_y());
-
-	/* set number of samples */
-	session->tile_manager.set_samples(session_params.samples);
-	session->reset(buffer_params, session_params.samples);
-	session->update_scene();
-
-	/* find object index. todo: is arbitrary - copied from mesh_displace.cpp */
-	for(size_t i = 0; i < scene->objects.size(); i++) {
-		if(strcmp(scene->objects[i]->name.c_str(), b_object.name().c_str()) == 0) {
-			object_index = i;
-			tri_offset = scene->objects[i]->mesh->tri_offset;
-			break;
-		}
+	if(!session->progress.get_cancel()) {
+		/* update scene */
+		BL::Object b_camera_override(b_engine.camera_override());
+		sync->sync_camera(b_render, b_camera_override, width, height, "");
+		sync->sync_data(b_render,
+						b_v3d,
+						b_camera_override,
+						width, height,
+						&python_thread_state,
+						b_rlay_name.c_str());
 	}
 
-	int object = object_index;
+	BakeData *bake_data = NULL;
 
-	BakeData *bake_data = scene->bake_manager->init(object, tri_offset, num_pixels);
+	if(!session->progress.get_cancel()) {
+		/* get buffer parameters */
+		SessionParams session_params = BlenderSync::get_session_params(b_engine, b_userpref, b_scene, background);
+		BufferParams buffer_params = BlenderSync::get_buffer_params(b_render, b_v3d, b_rv3d, scene->camera, width, height);
 
-	populate_bake_data(bake_data, object_id, pixel_array, num_pixels);
+		scene->bake_manager->set_shader_limit((size_t)b_engine.tile_x(), (size_t)b_engine.tile_y());
 
-	/* set number of samples */
-	session->tile_manager.set_samples(session_params.samples);
-	session->reset(buffer_params, session_params.samples);
-	session->update_scene();
+		/* set number of samples */
+		session->tile_manager.set_samples(session_params.samples);
+		session->reset(buffer_params, session_params.samples);
+		session->update_scene();
 
-	session->progress.set_update_callback(function_bind(&BlenderSession::update_bake_progress, this));
+		/* find object index. todo: is arbitrary - copied from mesh_displace.cpp */
+		size_t object_index = OBJECT_NONE;
+		int tri_offset = 0;
 
-	scene->bake_manager->bake(scene->device, &scene->dscene, scene, session->progress, shader_type, bake_pass_filter, bake_data, result);
+		for(size_t i = 0; i < scene->objects.size(); i++) {
+			if(strcmp(scene->objects[i]->name.c_str(), b_object.name().c_str()) == 0) {
+				object_index = i;
+				tri_offset = scene->objects[i]->mesh->tri_offset;
+				break;
+			}
+		}
+
+		int object = object_index;
+
+		bake_data = scene->bake_manager->init(object, tri_offset, num_pixels);
+		populate_bake_data(bake_data, object_id, pixel_array, num_pixels);
+
+		/* set number of samples */
+		session->tile_manager.set_samples(session_params.samples);
+		session->reset(buffer_params, session_params.samples);
+		session->update_scene();
+
+		session->progress.set_update_callback(function_bind(&BlenderSession::update_bake_progress, this));
+	}
+
+	/* Perform bake. Check cancel to avoid crash with incomplete scene data. */
+	if(!session->progress.get_cancel()) {
+		scene->bake_manager->bake(scene->device, &scene->dscene, scene, session->progress, shader_type, bake_pass_filter, bake_data, result);
+	}
 
 	/* free all memory used (host and device), so we wouldn't leave render
 	 * engine with extra memory allocated
@@ -1013,7 +1013,8 @@ void BlenderSession::builtin_image_info(const string &builtin_name,
                                         int &width,
                                         int &height,
                                         int &depth,
-                                        int &channels)
+                                        int &channels,
+                                        bool& free_cache)
 {
 	/* empty image */
 	is_float = false;
@@ -1021,6 +1022,7 @@ void BlenderSession::builtin_image_info(const string &builtin_name,
 	height = 1;
 	depth = 0;
 	channels = 0;
+	free_cache = false;
 
 	if(!builtin_data)
 		return;
@@ -1034,6 +1036,7 @@ void BlenderSession::builtin_image_info(const string &builtin_name,
 		/* image data */
 		BL::Image b_image(b_id);
 
+		free_cache = !b_image.has_data();
 		is_float = b_image.is_float();
 		width = b_image.size()[0];
 		height = b_image.size()[1];
@@ -1094,7 +1097,8 @@ void BlenderSession::builtin_image_info(const string &builtin_name,
 bool BlenderSession::builtin_image_pixels(const string &builtin_name,
                                           void *builtin_data,
                                           unsigned char *pixels,
-                                          const size_t pixels_size)
+                                          const size_t pixels_size,
+                                          const bool free_cache)
 {
 	if(!builtin_data) {
 		return false;
@@ -1115,7 +1119,6 @@ bool BlenderSession::builtin_image_pixels(const string &builtin_name,
 
 	if(image_pixels && num_pixels * channels == pixels_size) {
 		memcpy(pixels, image_pixels, pixels_size * sizeof(unsigned char));
-		MEM_freeN(image_pixels);
 	}
 	else {
 		if(channels == 1) {
@@ -1134,6 +1137,16 @@ bool BlenderSession::builtin_image_pixels(const string &builtin_name,
 			}
 		}
 	}
+
+	if(image_pixels) {
+		MEM_freeN(image_pixels);
+	}
+
+	/* Free image buffers to save memory during render. */
+	if(free_cache) {
+		b_image.buffers_free();
+	}
+
 	/* Premultiply, byte images are always straight for Blender. */
 	unsigned char *cp = pixels;
 	for(size_t i = 0; i < num_pixels; i++, cp += channels) {
@@ -1147,7 +1160,8 @@ bool BlenderSession::builtin_image_pixels(const string &builtin_name,
 bool BlenderSession::builtin_image_float_pixels(const string &builtin_name,
                                                 void *builtin_data,
                                                 float *pixels,
-                                                const size_t pixels_size)
+                                                const size_t pixels_size,
+                                                const bool free_cache)
 {
 	if(!builtin_data) {
 		return false;
@@ -1172,7 +1186,6 @@ bool BlenderSession::builtin_image_float_pixels(const string &builtin_name,
 
 		if(image_pixels && num_pixels * channels == pixels_size) {
 			memcpy(pixels, image_pixels, pixels_size * sizeof(float));
-			MEM_freeN(image_pixels);
 		}
 		else {
 			if(channels == 1) {
@@ -1192,6 +1205,15 @@ bool BlenderSession::builtin_image_float_pixels(const string &builtin_name,
 			}
 		}
 
+		if(image_pixels) {
+			MEM_freeN(image_pixels);
+		}
+
+		/* Free image buffers to save memory during render. */
+		if(free_cache) {
+			b_image.buffers_free();
+		}
+
 		return true;
 	}
 	else if(b_id.is_a(&RNA_Object)) {

intern/cycles/blender/blender_session.h

@@ -156,15 +156,18 @@ protected:
 	                        int &width,
 	                        int &height,
 	                        int &depth,
-	                        int &channels);
+	                        int &channels,
+	                        bool &free_cache);
 	bool builtin_image_pixels(const string &builtin_name,
 	                          void *builtin_data,
 	                          unsigned char *pixels,
-	                          const size_t pixels_size);
+	                          const size_t pixels_size,
+	                          const bool free_cache);
 	bool builtin_image_float_pixels(const string &builtin_name,
 	                                void *builtin_data,
 	                                float *pixels,
-	                                const size_t pixels_size);
+	                                const size_t pixels_size,
+	                                const bool free_cache);
 
 	/* Update tile manager to reflect resumable render settings. */
 	void update_resumable_tile_manager(int num_samples);

intern/cycles/blender/blender_sync.cpp

@@ -329,6 +329,9 @@ void BlenderSync::sync_integrator()
 			integrator->ao_bounces = get_int(cscene, "ao_bounces_render");
 		}
 	}
+	else {
+		integrator->ao_bounces = 0;
+	}
 
 	if(integrator->modified(previntegrator))
 		integrator->tag_update(scene);
@@ -550,11 +553,16 @@ int BlenderSync::get_denoising_pass(BL::RenderPass& b_pass)
 }
 
 array<Pass> BlenderSync::sync_render_passes(BL::RenderLayer& b_rlay,
-                                            BL::SceneRenderLayer& b_srlay)
+                                            BL::SceneRenderLayer& b_srlay,
+                                            const SessionParams &session_params)
 {
 	array<Pass> passes;
 	Pass::add(PASS_COMBINED, passes);
 
+	if(!session_params.device.advanced_shading) {
+		return passes;
+	}
+
 	/* loop over passes */
 	BL::RenderLayer::passes_iterator b_pass_iter;
 
@@ -569,7 +577,9 @@ array<Pass> BlenderSync::sync_render_passes(BL::RenderLayer& b_rlay,
 	}
 
 	PointerRNA crp = RNA_pointer_get(&b_srlay.ptr, "cycles");
-	if(get_boolean(crp, "denoising_store_passes")) {
+	if(get_boolean(crp, "denoising_store_passes") &&
+	   get_boolean(crp, "use_denoising") &&
+	   !session_params.progressive_refine) {
 		b_engine.add_pass("Denoising Normal",          3, "XYZ", b_srlay.name().c_str());
 		b_engine.add_pass("Denoising Normal Variance", 3, "XYZ", b_srlay.name().c_str());
 		b_engine.add_pass("Denoising Albedo",          3, "RGB", b_srlay.name().c_str());
@@ -598,8 +608,6 @@ array<Pass> BlenderSync::sync_render_passes(BL::RenderLayer& b_rlay,
 		b_engine.add_pass("Debug Ray Bounces", 1, "X", b_srlay.name().c_str());
 		Pass::add(PASS_RAY_BOUNCES, passes);
 	}
-#else
-	(void) b_srlay;  /* Ignored. */
 #endif
 
 	return passes;

intern/cycles/blender/blender_sync.h

@@ -68,7 +68,8 @@ public:
 	               const char *layer = 0);
 	void sync_render_layers(BL::SpaceView3D& b_v3d, const char *layer);
 	array<Pass> sync_render_passes(BL::RenderLayer& b_rlay,
-	                               BL::SceneRenderLayer& b_srlay);
+	                               BL::SceneRenderLayer& b_srlay,
+	                               const SessionParams &session_params);
 	void sync_integrator();
 	void sync_camera(BL::RenderSettings& b_render,
 	                 BL::Object& b_override,

intern/cycles/blender/blender_util.h

@@ -51,8 +51,8 @@ static inline BL::Mesh object_to_mesh(BL::BlendData& data,
                                       bool calc_undeformed,
                                       Mesh::SubdivisionType subdivision_type)
 {
-	bool subsurf_mod_show_render;
-	bool subsurf_mod_show_viewport;
+	bool subsurf_mod_show_render = false;
+	bool subsurf_mod_show_viewport = false;
 
 	if(subdivision_type != Mesh::SUBDIVISION_NONE) {
 		BL::Modifier subsurf_mod = object.modifiers[object.modifiers.length()-1];
@@ -299,7 +299,7 @@ static inline uint get_layer(const BL::Array<int, 20>& array)
 	for(uint i = 0; i < 20; i++)
 		if(array[i])
 			layer |= (1 << i);
-	
+
 	return layer;
 }
 
@@ -434,7 +434,7 @@ static inline string get_string(PointerRNA& ptr, const char *name)
 	string str(cstr);
 	if(cstr != cstrbuf)
 		MEM_freeN(cstr);
-	
+
 	return str;
 }
 
@@ -451,7 +451,7 @@ static inline string blender_absolute_path(BL::BlendData& b_data,
 {
 	if(path.size() >= 2 && path[0] == '/' && path[1] == '/') {
 		string dirname;
-		
+
 		if(b_id.library()) {
 			BL::ID b_library_id(b_id.library());
 			dirname = blender_absolute_path(b_data,
@@ -544,7 +544,7 @@ static inline BL::SmokeDomainSettings object_smoke_domain_find(BL::Object& b_ob)
 				return b_smd.domain_settings();
 		}
 	}
-	
+
 	return BL::SmokeDomainSettings(PointerRNA_NULL);
 }
 
@@ -816,4 +816,3 @@ protected:
 CCL_NAMESPACE_END
 
 #endif /* __BLENDER_UTIL_H__ */
-

intern/cycles/bvh/bvh4.cpp

@@ -242,21 +242,23 @@ void BVH4::pack_unaligned_node(int idx,
 		 * so kernel might safely assume there are always 4 child nodes.
 		 */
 
-		data[1][i] = 1.0f;
+		const float inf = FLT_MAX / 1000.0f;
+
+		data[1][i] = inf;
 		data[2][i] = 0.0f;
 		data[3][i] = 0.0f;
 
 		data[4][i] = 0.0f;
-		data[5][i] = 0.0f;
+		data[5][i] = inf;
 		data[6][i] = 0.0f;
 
 		data[7][i] = 0.0f;
 		data[8][i] = 0.0f;
-		data[9][i] = 0.0f;
+		data[9][i] = inf;
 
-		data[10][i] = -FLT_MAX;
-		data[11][i] = -FLT_MAX;
-		data[12][i] = -FLT_MAX;
+		data[10][i] = -inf;
+		data[11][i] = -inf;
+		data[12][i] = -inf;
 
 		data[13][i] = __int_as_float(0);
 	}

intern/cycles/bvh/bvh_build.cpp

@@ -1040,7 +1040,6 @@ BVHNode* BVHBuild::create_leaf_node(const BVHRange& range,
 		 */
 		start_index = spatial_free_index;
 		spatial_free_index += range.size();
-
 		/* Extend an array when needed. */
 		const size_t range_end = start_index + range.size();
 		if(prim_type.size() < range_end) {
@@ -1066,8 +1065,6 @@ BVHNode* BVHBuild::create_leaf_node(const BVHRange& range,
 				prim_time.resize(range_end);
 			}
 		}
-		spatial_spin_lock.unlock();
-
 		/* Perform actual data copy. */
 		if(new_leaf_data_size > 0) {
 			memcpy(&prim_type[start_index], &local_prim_type[0], new_leaf_data_size);
@@ -1077,6 +1074,7 @@ BVHNode* BVHBuild::create_leaf_node(const BVHRange& range,
 				memcpy(&prim_time[start_index], &local_prim_time[0], sizeof(float2)*num_new_leaf_data);
 			}
 		}
+		spatial_spin_lock.unlock();
 	}
 	else {
 		/* For the regular BVH builder we simply copy new data starting at the

intern/cycles/device/device.cpp

@@ -68,6 +68,8 @@ std::ostream& operator <<(std::ostream &os,
 	   << string_from_bool(requested_features.use_transparent) << std::endl;
 	os << "Use Principled BSDF: "
 	   << string_from_bool(requested_features.use_principled) << std::endl;
+	os << "Use Denoising: "
+	   << string_from_bool(requested_features.use_denoising) << std::endl;
 	return os;
 }
 

intern/cycles/device/device.h

@@ -127,6 +127,9 @@ public:
 	/* Per-uber shader usage flags. */
 	bool use_principled;
 
+	/* Denoising features. */
+	bool use_denoising;
+
 	DeviceRequestedFeatures()
 	{
 		/* TODO(sergey): Find more meaningful defaults. */
@@ -145,6 +148,7 @@ public:
 		use_transparent = false;
 		use_shadow_tricks = false;
 		use_principled = false;
+		use_denoising = false;
 	}
 
 	bool modified(const DeviceRequestedFeatures& requested_features)
@@ -163,7 +167,8 @@ public:
 		         use_patch_evaluation == requested_features.use_patch_evaluation &&
 		         use_transparent == requested_features.use_transparent &&
 		         use_shadow_tricks == requested_features.use_shadow_tricks &&
-		         use_principled == requested_features.use_principled);
+		         use_principled == requested_features.use_principled &&
+		         use_denoising == requested_features.use_denoising);
 	}
 
 	/* Convert the requested features structure to a build options,
@@ -213,6 +218,9 @@ public:
 		if(!use_principled) {
 			build_options += " -D__NO_PRINCIPLED__";
 		}
+		if(!use_denoising) {
+			build_options += " -D__NO_DENOISING__";
+		}
 		return build_options;
 	}
 };

intern/cycles/device/device_cpu.cpp

@@ -119,7 +119,7 @@ public:
 		}
 #endif
 
-		if(strstr(architecture_name, logged_architecture) != 0) {
+		if(strcmp(architecture_name, logged_architecture) != 0) {
 			VLOG(1) << "Will be using " << architecture_name << " kernels.";
 			logged_architecture = architecture_name;
 		}
@@ -149,7 +149,8 @@ public:
 	                                            device_memory& use_queues_flag,
 	                                            device_memory& work_pool_wgs);
 
-	virtual SplitKernelFunction* get_split_kernel_function(string kernel_name, const DeviceRequestedFeatures&);
+	virtual SplitKernelFunction* get_split_kernel_function(const string& kernel_name,
+	                                                       const DeviceRequestedFeatures&);
 	virtual int2 split_kernel_local_size();
 	virtual int2 split_kernel_global_size(device_memory& kg, device_memory& data, DeviceTask *task);
 	virtual uint64_t state_buffer_size(device_memory& kg, device_memory& data, size_t num_threads);
@@ -176,6 +177,7 @@ public:
 
 	KernelFunctions<void(*)(int, TilesInfo*, int, int, float*, float*, float*, float*, float*, int*, int, int, bool)> filter_divide_shadow_kernel;
 	KernelFunctions<void(*)(int, TilesInfo*, int, int, int, int, float*, float*, int*, int, int, bool)>               filter_get_feature_kernel;
+	KernelFunctions<void(*)(int, int, float*, float*, float*, float*, int*, int)>                                     filter_detect_outliers_kernel;
 	KernelFunctions<void(*)(int, int, float*, float*, float*, float*, int*, int)>                                     filter_combine_halves_kernel;
 
 	KernelFunctions<void(*)(int, int, float*, float*, float*, int*, int, int, float, float)> filter_nlm_calc_difference_kernel;
@@ -184,9 +186,9 @@ public:
 	KernelFunctions<void(*)(int, int, float*, float*, float*, float*, int*, int, int)>       filter_nlm_update_output_kernel;
 	KernelFunctions<void(*)(float*, float*, int*, int)>                                      filter_nlm_normalize_kernel;
 
-	KernelFunctions<void(*)(float*, int, int, int, float*, int*, int*, int, int, float)>                                              filter_construct_transform_kernel;
-	KernelFunctions<void(*)(int, int, float*, float*, float*, float*, float*, int*, float*, float3*, int*, int*, int, int, int, int)> filter_nlm_construct_gramian_kernel;
-	KernelFunctions<void(*)(int, int, int, int, int, float*, int*, float*, float3*, int*, int)>                                       filter_finalize_kernel;
+	KernelFunctions<void(*)(float*, int, int, int, float*, int*, int*, int, int, float)>                              filter_construct_transform_kernel;
+	KernelFunctions<void(*)(int, int, float*, float*, float*, int*, float*, float3*, int*, int*, int, int, int, int)> filter_nlm_construct_gramian_kernel;
+	KernelFunctions<void(*)(int, int, int, int, int, float*, int*, float*, float3*, int*, int)>                       filter_finalize_kernel;
 
 	KernelFunctions<void(*)(KernelGlobals *, ccl_constant KernelData*, ccl_global void*, int, ccl_global char*,
 	                       ccl_global uint*, int, int, int, int, int, int, int, int, ccl_global int*, int,
@@ -210,6 +212,7 @@ public:
 	  REGISTER_KERNEL(shader),
 	  REGISTER_KERNEL(filter_divide_shadow),
 	  REGISTER_KERNEL(filter_get_feature),
+	  REGISTER_KERNEL(filter_detect_outliers),
 	  REGISTER_KERNEL(filter_combine_halves),
 	  REGISTER_KERNEL(filter_nlm_calc_difference),
 	  REGISTER_KERNEL(filter_nlm_blur),
@@ -246,6 +249,7 @@ public:
 		REGISTER_SPLIT_KERNEL(direct_lighting);
 		REGISTER_SPLIT_KERNEL(shadow_blocked_ao);
 		REGISTER_SPLIT_KERNEL(shadow_blocked_dl);
+		REGISTER_SPLIT_KERNEL(enqueue_inactive);
 		REGISTER_SPLIT_KERNEL(next_iteration_setup);
 		REGISTER_SPLIT_KERNEL(indirect_subsurface);
 		REGISTER_SPLIT_KERNEL(buffer_update);
@@ -463,8 +467,6 @@ public:
 
 	bool denoising_reconstruct(device_ptr color_ptr,
 	                           device_ptr color_variance_ptr,
-	                           device_ptr guide_ptr,
-	                           device_ptr guide_variance_ptr,
 	                           device_ptr output_ptr,
 	                           DenoisingTask *task)
 	{
@@ -483,8 +485,8 @@ public:
 			                     task->reconstruction_state.source_w - max(0, dx),
 			                     task->reconstruction_state.source_h - max(0, dy)};
 			filter_nlm_calc_difference_kernel()(dx, dy,
-			                                    (float*) guide_ptr,
-			                                    (float*) guide_variance_ptr,
+			                                    (float*) color_ptr,
+			                                    (float*) color_variance_ptr,
 			                                    difference,
 			                                    local_rect,
 			                                    task->buffer.w,
@@ -497,8 +499,6 @@ public:
 			filter_nlm_construct_gramian_kernel()(dx, dy,
 			                                      blurDifference,
 			                                      (float*)  task->buffer.mem.device_pointer,
-			                                      (float*)  color_ptr,
-			                                      (float*)  color_variance_ptr,
 			                                      (float*)  task->storage.transform.device_pointer,
 			                                      (int*)    task->storage.rank.device_pointer,
 			                                      (float*)  task->storage.XtWX.device_pointer,
@@ -530,9 +530,8 @@ public:
 
 	bool denoising_combine_halves(device_ptr a_ptr, device_ptr b_ptr,
 	                              device_ptr mean_ptr, device_ptr variance_ptr,
-	                              int r, int4 rect, DenoisingTask *task)
+	                              int r, int4 rect, DenoisingTask * /*task*/)
 	{
-		(void) task;
 		for(int y = rect.y; y < rect.w; y++) {
 			for(int x = rect.x; x < rect.z; x++) {
 				filter_combine_halves_kernel()(x, y,
@@ -594,6 +593,26 @@ public:
 		return true;
 	}
 
+	bool denoising_detect_outliers(device_ptr image_ptr,
+	                               device_ptr variance_ptr,
+	                               device_ptr depth_ptr,
+	                               device_ptr output_ptr,
+	                               DenoisingTask *task)
+	{
+		for(int y = task->rect.y; y < task->rect.w; y++) {
+			for(int x = task->rect.x; x < task->rect.z; x++) {
+				filter_detect_outliers_kernel()(x, y,
+				                                (float*) image_ptr,
+				                                (float*) variance_ptr,
+				                                (float*) depth_ptr,
+				                                (float*) output_ptr,
+				                                &task->rect.x,
+				                                task->buffer.pass_stride);
+			}
+		}
+		return true;
+	}
+
 	void path_trace(DeviceTask &task, RenderTile &tile, KernelGlobals *kg)
 	{
 		float *render_buffer = (float*)tile.buffer;
@@ -627,11 +646,12 @@ public:
 		DenoisingTask denoising(this);
 
 		denoising.functions.construct_transform = function_bind(&CPUDevice::denoising_construct_transform, this, &denoising);
-		denoising.functions.reconstruct = function_bind(&CPUDevice::denoising_reconstruct, this, _1, _2, _3, _4, _5, &denoising);
+		denoising.functions.reconstruct = function_bind(&CPUDevice::denoising_reconstruct, this, _1, _2, _3, &denoising);
 		denoising.functions.divide_shadow = function_bind(&CPUDevice::denoising_divide_shadow, this, _1, _2, _3, _4, _5, &denoising);
 		denoising.functions.non_local_means = function_bind(&CPUDevice::denoising_non_local_means, this, _1, _2, _3, _4, &denoising);
 		denoising.functions.combine_halves = function_bind(&CPUDevice::denoising_combine_halves, this, _1, _2, _3, _4, _5, _6, &denoising);
 		denoising.functions.get_feature = function_bind(&CPUDevice::denoising_get_feature, this, _1, _2, _3, _4, &denoising);
+		denoising.functions.detect_outliers = function_bind(&CPUDevice::denoising_detect_outliers, this, _1, _2, _3, _4, &denoising);
 		denoising.functions.set_tiles = function_bind(&CPUDevice::denoising_set_tiles, this, _1, &denoising);
 
 		denoising.filter_area = make_int4(tile.x, tile.y, tile.w, tile.h);
@@ -663,7 +683,7 @@ public:
 		kgbuffer.resize(1);
 		mem_alloc("kernel_globals", kgbuffer, MEM_READ_WRITE);
 
-		KernelGlobals *kg = new((void*)kgbuffer.device_pointer) KernelGlobals(thread_kernel_globals_init());
+		KernelGlobals *kg = new ((void*) kgbuffer.device_pointer) KernelGlobals(thread_kernel_globals_init());
 
 		CPUSplitKernel *split_kernel = NULL;
 		if(use_split_kernel) {
@@ -702,6 +722,7 @@ public:
 		}
 
 		thread_kernel_globals_free((KernelGlobals*)kgbuffer.device_pointer);
+		kg->~KernelGlobals();
 		mem_free(kgbuffer);
 		delete split_kernel;
 	}
@@ -912,7 +933,8 @@ bool CPUSplitKernel::enqueue_split_kernel_data_init(const KernelDimensions& dim,
 	return true;
 }
 
-SplitKernelFunction* CPUSplitKernel::get_split_kernel_function(string kernel_name, const DeviceRequestedFeatures&)
+SplitKernelFunction* CPUSplitKernel::get_split_kernel_function(const string& kernel_name,
+                                                               const DeviceRequestedFeatures&)
 {
 	CPUSplitKernelFunction *kernel = new CPUSplitKernelFunction(device);
 

intern/cycles/device/device_cuda.cpp

@@ -105,7 +105,8 @@ public:
 	                                            device_memory& use_queues_flag,
 	                                            device_memory& work_pool_wgs);
 
-	virtual SplitKernelFunction* get_split_kernel_function(string kernel_name, const DeviceRequestedFeatures&);
+	virtual SplitKernelFunction* get_split_kernel_function(const string& kernel_name,
+	                                                       const DeviceRequestedFeatures&);
 	virtual int2 split_kernel_local_size();
 	virtual int2 split_kernel_global_size(device_memory& kg, device_memory& data, DeviceTask *task);
 };
@@ -225,6 +226,9 @@ public:
 		cuDevice = 0;
 		cuContext = 0;
 
+		cuModule = 0;
+		cuFilterModule = 0;
+
 		split_kernel = NULL;
 
 		need_bindless_mapping = false;
@@ -487,6 +491,16 @@ public:
 
 	bool load_kernels(const DeviceRequestedFeatures& requested_features)
 	{
+		/* TODO(sergey): Support kernels re-load for CUDA devices.
+		 *
+		 * Currently re-loading kernel will invalidate memory pointers,
+		 * causing problems in cuCtxSynchronize.
+		 */
+		if(cuFilterModule && cuModule) {
+			VLOG(1) << "Skipping kernel reload, not currently supported.";
+			return true;
+		}
+
 		/* check if cuda init succeeded */
 		if(cuContext == 0)
 			return false;
@@ -949,7 +963,7 @@ public:
 		cuda_push_context();
 
 		int4 rect = task->rect;
-		int w = rect.z-rect.x;
+		int w = align_up(rect.z-rect.x, 4);
 		int h = rect.w-rect.y;
 		int r = task->nlm_state.r;
 		int f = task->nlm_state.f;
@@ -1038,8 +1052,6 @@ public:
 
 	bool denoising_reconstruct(device_ptr color_ptr,
 	                           device_ptr color_variance_ptr,
-	                           device_ptr guide_ptr,
-	                           device_ptr guide_variance_ptr,
 	                           device_ptr output_ptr,
 	                           DenoisingTask *task)
 	{
@@ -1083,8 +1095,8 @@ public:
 			                     task->reconstruction_state.source_h - max(0, dy)};
 
 			void *calc_difference_args[] = {&dx, &dy,
-			                                &guide_ptr,
-			                                &guide_variance_ptr,
+			                                &color_ptr,
+			                                &color_variance_ptr,
 			                                &difference,
 			                                &local_rect,
 			                                &task->buffer.w,
@@ -1113,8 +1125,6 @@ public:
 			void *construct_gramian_args[] = {&dx, &dy,
 			                                  &blurDifference,
 			                                  &task->buffer.mem.device_pointer,
-			                                  &color_ptr,
-			                                  &color_variance_ptr,
 			                                  &task->storage.transform.device_pointer,
 			                                  &task->storage.rank.device_pointer,
 			                                  &task->storage.XtWX.device_pointer,
@@ -1148,8 +1158,6 @@ public:
 	                              device_ptr mean_ptr, device_ptr variance_ptr,
 	                              int r, int4 rect, DenoisingTask *task)
 	{
-		(void) task;
-
 		if(have_error())
 			return false;
 
@@ -1179,8 +1187,6 @@ public:
 	                             device_ptr sample_variance_ptr, device_ptr sv_variance_ptr,
 	                             device_ptr buffer_variance_ptr, DenoisingTask *task)
 	{
-		(void) task;
-
 		if(have_error())
 			return false;
 
@@ -1248,16 +1254,49 @@ public:
 		return !have_error();
 	}
 
+	bool denoising_detect_outliers(device_ptr image_ptr,
+	                               device_ptr variance_ptr,
+	                               device_ptr depth_ptr,
+	                               device_ptr output_ptr,
+	                               DenoisingTask *task)
+	{
+		if(have_error())
+			return false;
+
+		cuda_push_context();
+
+		CUfunction cuFilterDetectOutliers;
+		cuda_assert(cuModuleGetFunction(&cuFilterDetectOutliers, cuFilterModule, "kernel_cuda_filter_detect_outliers"));
+		cuda_assert(cuFuncSetCacheConfig(cuFilterDetectOutliers, CU_FUNC_CACHE_PREFER_L1));
+		CUDA_GET_BLOCKSIZE(cuFilterDetectOutliers,
+		                   task->rect.z-task->rect.x,
+		                   task->rect.w-task->rect.y);
+
+		void *args[] = {&image_ptr,
+		                &variance_ptr,
+		                &depth_ptr,
+		                &output_ptr,
+		                &task->rect,
+		                &task->buffer.pass_stride};
+
+		CUDA_LAUNCH_KERNEL(cuFilterDetectOutliers, args);
+		cuda_assert(cuCtxSynchronize());
+
+		cuda_pop_context();
+		return !have_error();
+	}
+
 	void denoise(RenderTile &rtile, const DeviceTask &task)
 	{
 		DenoisingTask denoising(this);
 
 		denoising.functions.construct_transform = function_bind(&CUDADevice::denoising_construct_transform, this, &denoising);
-		denoising.functions.reconstruct = function_bind(&CUDADevice::denoising_reconstruct, this, _1, _2, _3, _4, _5, &denoising);
+		denoising.functions.reconstruct = function_bind(&CUDADevice::denoising_reconstruct, this, _1, _2, _3, &denoising);
 		denoising.functions.divide_shadow = function_bind(&CUDADevice::denoising_divide_shadow, this, _1, _2, _3, _4, _5, &denoising);
 		denoising.functions.non_local_means = function_bind(&CUDADevice::denoising_non_local_means, this, _1, _2, _3, _4, &denoising);
 		denoising.functions.combine_halves = function_bind(&CUDADevice::denoising_combine_halves, this, _1, _2, _3, _4, _5, _6, &denoising);
 		denoising.functions.get_feature = function_bind(&CUDADevice::denoising_get_feature, this, _1, _2, _3, _4, &denoising);
+		denoising.functions.detect_outliers = function_bind(&CUDADevice::denoising_detect_outliers, this, _1, _2, _3, _4, &denoising);
 		denoising.functions.set_tiles = function_bind(&CUDADevice::denoising_set_tiles, this, _1, &denoising);
 
 		denoising.filter_area = make_int4(rtile.x, rtile.y, rtile.w, rtile.h);
@@ -1999,7 +2038,8 @@ bool CUDASplitKernel::enqueue_split_kernel_data_init(const KernelDimensions& dim
 	return !device->have_error();
 }
 
-SplitKernelFunction* CUDASplitKernel::get_split_kernel_function(string kernel_name, const DeviceRequestedFeatures&)
+SplitKernelFunction* CUDASplitKernel::get_split_kernel_function(const string& kernel_name,
+                                                                const DeviceRequestedFeatures&)
 {
 	CUfunction func;
 

intern/cycles/device/device_denoising.cpp

@@ -139,9 +139,9 @@ bool DenoisingTask::run_denoising()
 		nlm_state.temporary_2_ptr = *nlm_temporary_2;
 		nlm_state.temporary_3_ptr = *nlm_temporary_3;
 
-		int mean_from[]     = { 0, 1, 2, 6,  7,  8, 12 };
-		int variance_from[] = { 3, 4, 5, 9, 10, 11, 13 };
-		int pass_to[]       = { 1, 2, 3, 0,  5,  6,  7 };
+		int mean_from[]     = { 0, 1, 2, 12, 6,  7, 8 };
+		int variance_from[] = { 3, 4, 5, 13, 9, 10, 11};
+		int pass_to[]       = { 1, 2, 3, 0,  5,  6,  7};
 		for(int pass = 0; pass < 7; pass++) {
 			device_sub_ptr feature_pass(device, buffer.mem, pass_to[pass]*buffer.pass_stride, buffer.pass_stride, MEM_READ_WRITE);
 			/* Get the unfiltered pass and its variance from the RenderBuffers. */
@@ -159,11 +159,25 @@ bool DenoisingTask::run_denoising()
 		int mean_to[]       = { 8,  9, 10};
 		int variance_to[]   = {11, 12, 13};
 		int num_color_passes = 3;
+
+		device_only_memory<float> temp_color;
+		temp_color.resize(3*buffer.pass_stride);
+		device->mem_alloc("Denoising temporary color", temp_color, MEM_READ_WRITE);
+
 		for(int pass = 0; pass < num_color_passes; pass++) {
-			device_sub_ptr color_pass    (device, buffer.mem,     mean_to[pass]*buffer.pass_stride, buffer.pass_stride, MEM_READ_WRITE);
+			device_sub_ptr color_pass(device, temp_color, pass*buffer.pass_stride, buffer.pass_stride, MEM_READ_WRITE);
 			device_sub_ptr color_var_pass(device, buffer.mem, variance_to[pass]*buffer.pass_stride, buffer.pass_stride, MEM_READ_WRITE);
 			functions.get_feature(mean_from[pass], variance_from[pass], *color_pass, *color_var_pass);
 		}
+
+		{
+			device_sub_ptr depth_pass    (device, buffer.mem,                                 0,   buffer.pass_stride, MEM_READ_WRITE);
+			device_sub_ptr color_var_pass(device, buffer.mem, variance_to[0]*buffer.pass_stride, 3*buffer.pass_stride, MEM_READ_WRITE);
+			device_sub_ptr output_pass   (device, buffer.mem,     mean_to[0]*buffer.pass_stride, 3*buffer.pass_stride, MEM_READ_WRITE);
+			functions.detect_outliers(temp_color.device_pointer, *color_var_pass, *depth_pass, *output_pass);
+		}
+
+		device->mem_free(temp_color);
 	}
 
 	storage.w = filter_area.z;
@@ -201,7 +215,7 @@ bool DenoisingTask::run_denoising()
 	{
 		device_sub_ptr color_ptr    (device, buffer.mem,  8*buffer.pass_stride, 3*buffer.pass_stride, MEM_READ_WRITE);
 		device_sub_ptr color_var_ptr(device, buffer.mem, 11*buffer.pass_stride, 3*buffer.pass_stride, MEM_READ_WRITE);
-		functions.reconstruct(*color_ptr, *color_var_ptr, *color_ptr, *color_var_ptr, render_buffer.ptr);
+		functions.reconstruct(*color_ptr, *color_var_ptr, render_buffer.ptr);
 	}
 
 	device->mem_free(storage.XtWX);

intern/cycles/device/device_denoising.h

@@ -58,8 +58,6 @@ public:
 		              )> non_local_means;
 		function<bool(device_ptr color_ptr,
 		              device_ptr color_variance_ptr,
-		              device_ptr guide_ptr,
-		              device_ptr guide_variance_ptr,
 		              device_ptr output_ptr
 		              )> reconstruct;
 		function<bool()> construct_transform;
@@ -82,6 +80,11 @@ public:
 		              device_ptr mean_ptr,
 		              device_ptr variance_ptr
 		              )> get_feature;
+		function<bool(device_ptr image_ptr,
+		              device_ptr variance_ptr,
+		              device_ptr depth_ptr,
+		              device_ptr output_ptr
+		              )> detect_outliers;
 		function<bool(device_ptr*)> set_tiles;
 	} functions;
 

intern/cycles/device/device_multi.cpp

@@ -335,7 +335,7 @@ public:
 				device_ptr original_ptr = mem.device_pointer;
 				mem.device_pointer = tiles[i].buffer;
 
-				/* Copy denoised tile to the host device. */
+				/* Copy denoised tile to the host. */
 				if(i == 4) {
 					tiles[i].buffers->copy_from_device(sub_device);
 				}

intern/cycles/device/device_opencl.cpp

@@ -130,10 +130,22 @@ string device_opencl_capabilities(void)
 		opencl_assert(func(id, what, sizeof(data), &data, NULL)); \
 		result += string_printf("%s: %s\n", name, data); \
 	} while(false)
+#define APPEND_STRING_EXTENSION_INFO(func, id, name, what) \
+	do { \
+		char data[1024] = "\0"; \
+		size_t length = 0; \
+		if(func(id, what, sizeof(data), &data, &length) == CL_SUCCESS) { \
+			if(length != 0 && data[0] != '\0') { \
+				result += string_printf("%s: %s\n", name, data); \
+			} \
+		} \
+	} while(false)
 #define APPEND_PLATFORM_STRING_INFO(id, name, what) \
 	APPEND_STRING_INFO(clGetPlatformInfo, id, "\tPlatform " name, what)
 #define APPEND_DEVICE_STRING_INFO(id, name, what) \
 	APPEND_STRING_INFO(clGetDeviceInfo, id, "\t\t\tDevice " name, what)
+#define APPEND_DEVICE_STRING_EXTENSION_INFO(id, name, what) \
+	APPEND_STRING_EXTENSION_INFO(clGetDeviceInfo, id, "\t\t\tDevice " name, what)
 
 	vector<cl_device_id> device_ids;
 	for(cl_uint platform = 0; platform < num_platforms; ++platform) {
@@ -167,6 +179,7 @@ string device_opencl_capabilities(void)
 			result += string_printf("\t\tDevice: #%u\n", device);
 
 			APPEND_DEVICE_STRING_INFO(device_id, "Name", CL_DEVICE_NAME);
+			APPEND_DEVICE_STRING_EXTENSION_INFO(device_id, "Board Name", CL_DEVICE_BOARD_NAME_AMD);
 			APPEND_DEVICE_STRING_INFO(device_id, "Vendor", CL_DEVICE_VENDOR);
 			APPEND_DEVICE_STRING_INFO(device_id, "OpenCL C Version", CL_DEVICE_OPENCL_C_VERSION);
 			APPEND_DEVICE_STRING_INFO(device_id, "Profile", CL_DEVICE_PROFILE);

intern/cycles/device/device_split_kernel.cpp

@@ -47,6 +47,7 @@ DeviceSplitKernel::DeviceSplitKernel(Device *device) : device(device)
 	kernel_direct_lighting = NULL;
 	kernel_shadow_blocked_ao = NULL;
 	kernel_shadow_blocked_dl = NULL;
+	kernel_enqueue_inactive = NULL;
 	kernel_next_iteration_setup = NULL;
 	kernel_indirect_subsurface = NULL;
 	kernel_buffer_update = NULL;
@@ -74,6 +75,7 @@ DeviceSplitKernel::~DeviceSplitKernel()
 	delete kernel_direct_lighting;
 	delete kernel_shadow_blocked_ao;
 	delete kernel_shadow_blocked_dl;
+	delete kernel_enqueue_inactive;
 	delete kernel_next_iteration_setup;
 	delete kernel_indirect_subsurface;
 	delete kernel_buffer_update;
@@ -101,6 +103,7 @@ bool DeviceSplitKernel::load_kernels(const DeviceRequestedFeatures& requested_fe
 	LOAD_KERNEL(direct_lighting);
 	LOAD_KERNEL(shadow_blocked_ao);
 	LOAD_KERNEL(shadow_blocked_dl);
+	LOAD_KERNEL(enqueue_inactive);
 	LOAD_KERNEL(next_iteration_setup);
 	LOAD_KERNEL(indirect_subsurface);
 	LOAD_KERNEL(buffer_update);
@@ -256,6 +259,7 @@ bool DeviceSplitKernel::path_trace(DeviceTask *task,
 				ENQUEUE_SPLIT_KERNEL(direct_lighting, global_size, local_size);
 				ENQUEUE_SPLIT_KERNEL(shadow_blocked_ao, global_size, local_size);
 				ENQUEUE_SPLIT_KERNEL(shadow_blocked_dl, global_size, local_size);
+				ENQUEUE_SPLIT_KERNEL(enqueue_inactive, global_size, local_size);
 				ENQUEUE_SPLIT_KERNEL(next_iteration_setup, global_size, local_size);
 				ENQUEUE_SPLIT_KERNEL(indirect_subsurface, global_size, local_size);
 				ENQUEUE_SPLIT_KERNEL(queue_enqueue, global_size, local_size);

intern/cycles/device/device_split_kernel.h

@@ -69,6 +69,7 @@ private:
 	SplitKernelFunction *kernel_direct_lighting;
 	SplitKernelFunction *kernel_shadow_blocked_ao;
 	SplitKernelFunction *kernel_shadow_blocked_dl;
+	SplitKernelFunction *kernel_enqueue_inactive;
 	SplitKernelFunction *kernel_next_iteration_setup;
 	SplitKernelFunction *kernel_indirect_subsurface;
 	SplitKernelFunction *kernel_buffer_update;
@@ -124,7 +125,8 @@ public:
 	                                            device_memory& use_queues_flag,
 	                                            device_memory& work_pool_wgs) = 0;
 
-	virtual SplitKernelFunction* get_split_kernel_function(string kernel_name, const DeviceRequestedFeatures&) = 0;
+	virtual SplitKernelFunction* get_split_kernel_function(const string& kernel_name,
+	                                                       const DeviceRequestedFeatures&) = 0;
 	virtual int2 split_kernel_local_size() = 0;
 	virtual int2 split_kernel_global_size(device_memory& kg, device_memory& data, DeviceTask *task) = 0;
 };

intern/cycles/device/opencl/opencl.h

@@ -27,6 +27,9 @@
 
 CCL_NAMESPACE_BEGIN
 
+/* Disable workarounds, seems to be working fine on latest drivers. */
+#define CYCLES_DISABLE_DRIVER_WORKAROUNDS
+
 /* Define CYCLES_DISABLE_DRIVER_WORKAROUNDS to disable workaounds for testing */
 #ifndef CYCLES_DISABLE_DRIVER_WORKAROUNDS
 /* Work around AMD driver hangs by ensuring each command is finished before doing anything else. */
@@ -84,7 +87,7 @@ public:
 	                                   string *error = NULL);
 	static bool device_version_check(cl_device_id device,
 	                                 string *error = NULL);
-	static string get_hardware_id(string platform_name,
+	static string get_hardware_id(const string& platform_name,
 	                              cl_device_id device_id);
 	static void get_usable_devices(vector<OpenCLPlatformDevice> *usable_devices,
 	                               bool force_all = false);
@@ -130,6 +133,11 @@ public:
 	                            cl_int* error = NULL);
 	static cl_device_type get_device_type(cl_device_id device_id);
 
+	static bool get_driver_version(cl_device_id device_id,
+	                               int *major,
+	                               int *minor,
+	                               cl_int* error = NULL);
+
 	static int mem_address_alignment(cl_device_id device_id);
 
 	/* Get somewhat more readable device name.
@@ -242,17 +250,17 @@ public:
 	public:
 		OpenCLProgram() : loaded(false), device(NULL) {}
 		OpenCLProgram(OpenCLDeviceBase *device,
-		              string program_name,
-		              string kernel_name,
-		              string kernel_build_options,
+		              const string& program_name,
+		              const string& kernel_name,
+		              const string& kernel_build_options,
 		              bool use_stdout = true);
 		~OpenCLProgram();
 
 		void add_kernel(ustring name);
 		void load();
 
-		bool is_loaded()    { return loaded; }
-		string get_log()    { return log; }
+		bool is_loaded() const { return loaded; }
+		const string& get_log() const { return log; }
 		void report_error();
 
 		cl_kernel operator()();
@@ -266,8 +274,8 @@ public:
 		bool load_binary(const string& clbin, const string *debug_src = NULL);
 		bool save_binary(const string& clbin);
 
-		void add_log(string msg, bool is_debug);
-		void add_error(string msg);
+		void add_log(const string& msg, bool is_debug);
+		void add_error(const string& msg);
 
 		bool loaded;
 		cl_program program;
@@ -390,8 +398,6 @@ protected:
 	bool denoising_construct_transform(DenoisingTask *task);
 	bool denoising_reconstruct(device_ptr color_ptr,
 	                           device_ptr color_variance_ptr,
-	                           device_ptr guide_ptr,
-	                           device_ptr guide_variance_ptr,
 	                           device_ptr output_ptr,
 	                           DenoisingTask *task);
 	bool denoising_combine_halves(device_ptr a_ptr,
@@ -411,6 +417,11 @@ protected:
 	                           device_ptr mean_ptr,
 	                           device_ptr variance_ptr,
 	                           DenoisingTask *task);
+	bool denoising_detect_outliers(device_ptr image_ptr,
+	                               device_ptr variance_ptr,
+	                               device_ptr depth_ptr,
+	                               device_ptr output_ptr,
+	                               DenoisingTask *task);
 	bool denoising_set_tiles(device_ptr *buffers,
 	                         DenoisingTask *task);
 

intern/cycles/device/opencl/opencl_base.cpp

@@ -20,6 +20,7 @@
 
 #include "kernel/kernel_types.h"
 
+#include "util/util_algorithm.h"
 #include "util/util_foreach.h"
 #include "util/util_logging.h"
 #include "util/util_md5.h"
@@ -216,6 +217,7 @@ bool OpenCLDeviceBase::load_kernels(const DeviceRequestedFeatures& requested_fea
 	denoising_program = OpenCLProgram(this, "denoising", "filter.cl", "");
 	denoising_program.add_kernel(ustring("filter_divide_shadow"));
 	denoising_program.add_kernel(ustring("filter_get_feature"));
+	denoising_program.add_kernel(ustring("filter_detect_outliers"));
 	denoising_program.add_kernel(ustring("filter_combine_halves"));
 	denoising_program.add_kernel(ustring("filter_construct_transform"));
 	denoising_program.add_kernel(ustring("filter_nlm_calc_difference"));
@@ -275,6 +277,25 @@ void OpenCLDeviceBase::mem_alloc(const char *name, device_memory& mem, MemoryTyp
 
 	size_t size = mem.memory_size();
 
+	/* check there is enough memory available for the allocation */
+	cl_ulong max_alloc_size = 0;
+	clGetDeviceInfo(cdDevice, CL_DEVICE_MAX_MEM_ALLOC_SIZE, sizeof(cl_ulong), &max_alloc_size, NULL);
+
+	if(DebugFlags().opencl.mem_limit) {
+		max_alloc_size = min(max_alloc_size,
+		                     cl_ulong(DebugFlags().opencl.mem_limit - stats.mem_used));
+	}
+
+	if(size > max_alloc_size) {
+		string error = "Scene too complex to fit in available memory.";
+		if(name != NULL) {
+			error += string_printf(" (allocating buffer %s failed.)", name);
+		}
+		set_error(error);
+
+		return;
+	}
+
 	cl_mem_flags mem_flag;
 	void *mem_ptr = NULL;
 
@@ -692,8 +713,6 @@ bool OpenCLDeviceBase::denoising_construct_transform(DenoisingTask *task)
 
 bool OpenCLDeviceBase::denoising_reconstruct(device_ptr color_ptr,
                                              device_ptr color_variance_ptr,
-                                             device_ptr guide_ptr,
-                                             device_ptr guide_variance_ptr,
                                              device_ptr output_ptr,
                                              DenoisingTask *task)
 {
@@ -702,8 +721,6 @@ bool OpenCLDeviceBase::denoising_reconstruct(device_ptr color_ptr,
 
 	cl_mem color_mem = CL_MEM_PTR(color_ptr);
 	cl_mem color_variance_mem = CL_MEM_PTR(color_variance_ptr);
-	cl_mem guide_mem = CL_MEM_PTR(guide_ptr);
-	cl_mem guide_variance_mem = CL_MEM_PTR(guide_variance_ptr);
 	cl_mem output_mem = CL_MEM_PTR(output_ptr);
 
 	cl_mem buffer_mem = CL_MEM_PTR(task->buffer.mem.device_pointer);
@@ -734,8 +751,8 @@ bool OpenCLDeviceBase::denoising_reconstruct(device_ptr color_ptr,
 
 		kernel_set_args(ckNLMCalcDifference, 0,
 		                dx, dy,
-		                guide_mem,
-		                guide_variance_mem,
+		                color_mem,
+		                color_variance_mem,
 		                difference,
 		                local_rect,
 		                task->buffer.w,
@@ -774,8 +791,6 @@ bool OpenCLDeviceBase::denoising_reconstruct(device_ptr color_ptr,
 		                dx, dy,
 		                blurDifference,
 		                buffer_mem,
-		                color_mem,
-		                color_variance_mem,
 		                transform_mem,
 		                rank_mem,
 		                XtWX_mem,
@@ -816,8 +831,6 @@ bool OpenCLDeviceBase::denoising_combine_halves(device_ptr a_ptr,
                                                 int r, int4 rect,
                                                 DenoisingTask *task)
 {
-	(void) task;
-
 	cl_mem a_mem = CL_MEM_PTR(a_ptr);
 	cl_mem b_mem = CL_MEM_PTR(b_ptr);
 	cl_mem mean_mem = CL_MEM_PTR(mean_ptr);
@@ -846,8 +859,6 @@ bool OpenCLDeviceBase::denoising_divide_shadow(device_ptr a_ptr,
                                                device_ptr buffer_variance_ptr,
                                                DenoisingTask *task)
 {
-	(void) task;
-
 	cl_mem a_mem = CL_MEM_PTR(a_ptr);
 	cl_mem b_mem = CL_MEM_PTR(b_ptr);
 	cl_mem sample_variance_mem = CL_MEM_PTR(sample_variance_ptr);
@@ -910,6 +921,33 @@ bool OpenCLDeviceBase::denoising_get_feature(int mean_offset,
 	return true;
 }
 
+bool OpenCLDeviceBase::denoising_detect_outliers(device_ptr image_ptr,
+                                                 device_ptr variance_ptr,
+                                                 device_ptr depth_ptr,
+                                                 device_ptr output_ptr,
+                                                 DenoisingTask *task)
+{
+	cl_mem image_mem = CL_MEM_PTR(image_ptr);
+	cl_mem variance_mem = CL_MEM_PTR(variance_ptr);
+	cl_mem depth_mem = CL_MEM_PTR(depth_ptr);
+	cl_mem output_mem = CL_MEM_PTR(output_ptr);
+
+	cl_kernel ckFilterDetectOutliers = denoising_program(ustring("filter_detect_outliers"));
+
+	kernel_set_args(ckFilterDetectOutliers, 0,
+	                image_mem,
+	                variance_mem,
+	                depth_mem,
+	                output_mem,
+	                task->rect,
+	                task->buffer.pass_stride);
+	enqueue_kernel(ckFilterDetectOutliers,
+	               task->rect.z-task->rect.x,
+	               task->rect.w-task->rect.y);
+
+	return true;
+}
+
 bool OpenCLDeviceBase::denoising_set_tiles(device_ptr *buffers,
                                            DenoisingTask *task)
 {
@@ -937,11 +975,12 @@ void OpenCLDeviceBase::denoise(RenderTile &rtile, const DeviceTask &task)
 
 	denoising.functions.set_tiles = function_bind(&OpenCLDeviceBase::denoising_set_tiles, this, _1, &denoising);
 	denoising.functions.construct_transform = function_bind(&OpenCLDeviceBase::denoising_construct_transform, this, &denoising);
-	denoising.functions.reconstruct = function_bind(&OpenCLDeviceBase::denoising_reconstruct, this, _1, _2, _3, _4, _5, &denoising);
+	denoising.functions.reconstruct = function_bind(&OpenCLDeviceBase::denoising_reconstruct, this, _1, _2, _3, &denoising);
 	denoising.functions.divide_shadow = function_bind(&OpenCLDeviceBase::denoising_divide_shadow, this, _1, _2, _3, _4, _5, &denoising);
 	denoising.functions.non_local_means = function_bind(&OpenCLDeviceBase::denoising_non_local_means, this, _1, _2, _3, _4, &denoising);
 	denoising.functions.combine_halves = function_bind(&OpenCLDeviceBase::denoising_combine_halves, this, _1, _2, _3, _4, _5, _6, &denoising);
 	denoising.functions.get_feature = function_bind(&OpenCLDeviceBase::denoising_get_feature, this, _1, _2, _3, _4, &denoising);
+	denoising.functions.detect_outliers = function_bind(&OpenCLDeviceBase::denoising_detect_outliers, this, _1, _2, _3, _4, &denoising);
 
 	denoising.filter_area = make_int4(rtile.x, rtile.y, rtile.w, rtile.h);
 	denoising.render_buffer.samples = rtile.sample;
@@ -1207,7 +1246,7 @@ void OpenCLDeviceBase::store_cached_kernel(
 }
 
 string OpenCLDeviceBase::build_options_for_base_program(
-        const DeviceRequestedFeatures& /*requested_features*/)
+        const DeviceRequestedFeatures& requested_features)
 {
 	/* TODO(sergey): By default we compile all features, meaning
 	 * mega kernel is not getting feature-based optimizations.
@@ -1215,6 +1254,14 @@ string OpenCLDeviceBase::build_options_for_base_program(
 	 * Ideally we need always compile kernel with as less features
 	 * enabled as possible to keep performance at it's max.
 	 */
+
+	/* For now disable baking when not in use as this has major
+	 * impact on kernel build times.
+	 */
+	if(!requested_features.use_baking) {
+		return "-D__NO_BAKING__";
+	}
+
 	return "";
 }
 

intern/cycles/device/opencl/opencl_split.cpp

@@ -25,6 +25,7 @@
 
 #include "device/device_split_kernel.h"
 
+#include "util/util_algorithm.h"
 #include "util/util_logging.h"
 #include "util/util_md5.h"
 #include "util/util_path.h"
@@ -70,6 +71,10 @@ public:
 		delete split_kernel;
 	}
 
+	virtual bool show_samples() const {
+		return true;
+	}
+
 	virtual bool load_kernels(const DeviceRequestedFeatures& requested_features,
 	                          vector<OpenCLDeviceBase::OpenCLProgram*> &programs)
 	{
@@ -172,17 +177,62 @@ protected:
 	friend class OpenCLSplitKernelFunction;
 };
 
+struct CachedSplitMemory {
+	int id;
+	device_memory *split_data;
+	device_memory *ray_state;
+	device_ptr *rng_state;
+	device_memory *queue_index;
+	device_memory *use_queues_flag;
+	device_memory *work_pools;
+	device_ptr *buffer;
+};
+
 class OpenCLSplitKernelFunction : public SplitKernelFunction {
 public:
 	OpenCLDeviceSplitKernel* device;
 	OpenCLDeviceBase::OpenCLProgram program;
+	CachedSplitMemory& cached_memory;
+	int cached_id;
 
-	OpenCLSplitKernelFunction(OpenCLDeviceSplitKernel* device) : device(device) {}
-	~OpenCLSplitKernelFunction() { program.release(); }
+	OpenCLSplitKernelFunction(OpenCLDeviceSplitKernel* device, CachedSplitMemory& cached_memory) :
+			device(device), cached_memory(cached_memory), cached_id(cached_memory.id-1)
+	{
+	}
+
+	~OpenCLSplitKernelFunction()
+	{
+		program.release();
+	}
 
 	virtual bool enqueue(const KernelDimensions& dim, device_memory& kg, device_memory& data)
 	{
-		device->kernel_set_args(program(), 0, kg, data);
+		if(cached_id != cached_memory.id) {
+			cl_uint start_arg_index =
+				device->kernel_set_args(program(),
+					            0,
+					            kg,
+					            data,
+					            *cached_memory.split_data,
+					            *cached_memory.ray_state,
+					            *cached_memory.rng_state);
+
+/* TODO(sergey): Avoid map lookup here. */
+#define KERNEL_TEX(type, ttype, name) \
+				device->set_kernel_arg_mem(program(), &start_arg_index, #name);
+#include "kernel/kernel_textures.h"
+#undef KERNEL_TEX
+
+			start_arg_index +=
+				device->kernel_set_args(program(),
+					            start_arg_index,
+					            *cached_memory.queue_index,
+					            *cached_memory.use_queues_flag,
+					            *cached_memory.work_pools,
+					            *cached_memory.buffer);
+
+			cached_id = cached_memory.id;
+		}
 
 		device->ciErr = clEnqueueNDRangeKernel(device->cqCommandQueue,
 		                                       program(),
@@ -209,14 +259,15 @@ public:
 
 class OpenCLSplitKernel : public DeviceSplitKernel {
 	OpenCLDeviceSplitKernel *device;
+	CachedSplitMemory cached_memory;
 public:
 	explicit OpenCLSplitKernel(OpenCLDeviceSplitKernel *device) : DeviceSplitKernel(device), device(device) {
 	}
 
-	virtual SplitKernelFunction* get_split_kernel_function(string kernel_name,
+	virtual SplitKernelFunction* get_split_kernel_function(const string& kernel_name,
 	                                                       const DeviceRequestedFeatures& requested_features)
 	{
-		OpenCLSplitKernelFunction* kernel = new OpenCLSplitKernelFunction(device);
+		OpenCLSplitKernelFunction* kernel = new OpenCLSplitKernelFunction(device, cached_memory);
 
 		bool single_program = OpenCLInfo::use_single_program();
 		kernel->program =
@@ -345,6 +396,15 @@ public:
 			return false;
 		}
 
+		cached_memory.split_data = &split_data;
+		cached_memory.ray_state = &ray_state;
+		cached_memory.rng_state = &rtile.rng_state;
+		cached_memory.queue_index = &queue_index;
+		cached_memory.use_queues_flag = &use_queues_flag;
+		cached_memory.work_pools = &work_pool_wgs;
+		cached_memory.buffer = &rtile.buffer;
+		cached_memory.id++;
+
 		return true;
 	}
 
@@ -364,12 +424,18 @@ public:
 
 		cl_ulong max_buffer_size;
 		clGetDeviceInfo(device->cdDevice, CL_DEVICE_MAX_MEM_ALLOC_SIZE, sizeof(cl_ulong), &max_buffer_size, NULL);
+
+		if(DebugFlags().opencl.mem_limit) {
+			max_buffer_size = min(max_buffer_size,
+			                      cl_ulong(DebugFlags().opencl.mem_limit - device->stats.mem_used));
+		}
+
 		VLOG(1) << "Maximum device allocation size: "
 		        << string_human_readable_number(max_buffer_size) << " bytes. ("
 		        << string_human_readable_size(max_buffer_size) << ").";
 
 		size_t num_elements = max_elements_for_max_buffer_size(kg, data, max_buffer_size / 2);
-		int2 global_size = make_int2(round_down((int)sqrt(num_elements), 64), (int)sqrt(num_elements));
+		int2 global_size = make_int2(max(round_down((int)sqrt(num_elements), 64), 64), (int)sqrt(num_elements));
 		VLOG(1) << "Global size: " << global_size << ".";
 		return global_size;
 	}

intern/cycles/device/opencl/opencl_util.cpp

@@ -241,9 +241,9 @@ string OpenCLCache::get_kernel_md5()
 }
 
 OpenCLDeviceBase::OpenCLProgram::OpenCLProgram(OpenCLDeviceBase *device,
-                                               string program_name,
-                                               string kernel_file,
-                                               string kernel_build_options,
+                                               const string& program_name,
+                                               const string& kernel_file,
+                                               const string& kernel_build_options,
                                                bool use_stdout)
  : device(device),
    program_name(program_name),
@@ -274,7 +274,7 @@ void OpenCLDeviceBase::OpenCLProgram::release()
 	}
 }
 
-void OpenCLDeviceBase::OpenCLProgram::add_log(string msg, bool debug)
+void OpenCLDeviceBase::OpenCLProgram::add_log(const string& msg, bool debug)
 {
 	if(!use_stdout) {
 		log += msg + "\n";
@@ -288,7 +288,7 @@ void OpenCLDeviceBase::OpenCLProgram::add_log(string msg, bool debug)
 	}
 }
 
-void OpenCLDeviceBase::OpenCLProgram::add_error(string msg)
+void OpenCLDeviceBase::OpenCLProgram::add_error(const string& msg)
 {
 	if(use_stdout) {
 		fprintf(stderr, "%s\n", msg.c_str());
@@ -608,6 +608,14 @@ bool OpenCLInfo::device_supported(const string& platform_name,
 	if(!get_device_name(device_id, &device_name)) {
 		return false;
 	}
+
+	int driver_major = 0;
+	int driver_minor = 0;
+	if(!get_driver_version(device_id, &driver_major, &driver_minor)) {
+		return false;
+	}
+	VLOG(3) << "OpenCL driver version " << driver_major << "." << driver_minor;
+
 	/* It is possible tyo have Iris GPU on AMD/Apple OpenCL framework
 	 * (aka, it will not be on Intel framework). This isn't supported
 	 * and needs an explicit blacklist.
@@ -618,6 +626,21 @@ bool OpenCLInfo::device_supported(const string& platform_name,
 	if(platform_name == "AMD Accelerated Parallel Processing" &&
 	   device_type == CL_DEVICE_TYPE_GPU)
 	{
+		if(driver_major < 2236) {
+			VLOG(1) << "AMD driver version " << driver_major << "." << driver_minor << " not supported.";
+			return false;
+		}
+		const char *blacklist[] = {
+			/* GCN 1 */
+			"Tahiti", "Pitcairn", "Capeverde", "Oland",
+			NULL
+		};
+		for (int i = 0; blacklist[i] != NULL; i++) {
+			if(device_name == blacklist[i]) {
+				VLOG(1) << "AMD device " << device_name << " not supported";
+				return false;
+			}
+		}
 		return true;
 	}
 	if(platform_name == "Apple" && device_type == CL_DEVICE_TYPE_GPU) {
@@ -684,7 +707,7 @@ bool OpenCLInfo::device_version_check(cl_device_id device,
 	return true;
 }
 
-string OpenCLInfo::get_hardware_id(string platform_name, cl_device_id device_id)
+string OpenCLInfo::get_hardware_id(const string& platform_name, cl_device_id device_id)
 {
 	if(platform_name == "AMD Accelerated Parallel Processing" || platform_name == "Apple") {
 		/* Use cl_amd_device_topology extension. */
@@ -1063,7 +1086,7 @@ string OpenCLInfo::get_readable_device_name(cl_device_id device_id)
 	                   CL_DEVICE_BOARD_NAME_AMD,
 	                   sizeof(board_name),
 	                   &board_name,
-					   &length) == CL_SUCCESS)
+	                   &length) == CL_SUCCESS)
 	{
 		if(length != 0 && board_name[0] != '\0') {
 			return board_name;
@@ -1073,6 +1096,34 @@ string OpenCLInfo::get_readable_device_name(cl_device_id device_id)
 	return get_device_name(device_id);
 }
 
+bool OpenCLInfo::get_driver_version(cl_device_id device_id,
+                                    int *major,
+                                    int *minor,
+                                    cl_int* error)
+{
+	char buffer[1024];
+	cl_int err;
+	if((err = clGetDeviceInfo(device_id,
+	                          CL_DRIVER_VERSION,
+	                          sizeof(buffer),
+	                          &buffer,
+	                          NULL)) != CL_SUCCESS)
+	{
+		if(error != NULL) {
+			*error = err;
+		}
+		return false;
+	}
+	if(error != NULL) {
+		*error = CL_SUCCESS;
+	}
+	if(sscanf(buffer, "%d.%d", major, minor) < 2) {
+		VLOG(1) << string_printf("OpenCL: failed to parse driver version string (%s).", buffer);
+		return false;
+	}
+	return true;
+}
+
 int OpenCLInfo::mem_address_alignment(cl_device_id device_id)
 {
 	int base_align_bits;

intern/cycles/kernel/CMakeLists.txt

@@ -45,6 +45,7 @@ set(SRC
 	kernels/opencl/kernel_direct_lighting.cl
 	kernels/opencl/kernel_shadow_blocked_ao.cl
 	kernels/opencl/kernel_shadow_blocked_dl.cl
+	kernels/opencl/kernel_enqueue_inactive.cl
 	kernels/opencl/kernel_next_iteration_setup.cl
 	kernels/opencl/kernel_indirect_subsurface.cl
 	kernels/opencl/kernel_buffer_update.cl
@@ -121,6 +122,10 @@ set(SRC_KERNELS_CUDA_HEADERS
 	kernels/cuda/kernel_config.h
 )
 
+set(SRC_KERNELS_OPENCL_HEADERS
+	kernels/opencl/kernel_split_function.h
+)
+
 set(SRC_CLOSURE_HEADERS
 	closure/alloc.h
 	closure/bsdf.h
@@ -278,6 +283,7 @@ set(SRC_SPLIT_HEADERS
 	split/kernel_data_init.h
 	split/kernel_direct_lighting.h
 	split/kernel_do_volume.h
+	split/kernel_enqueue_inactive.h
 	split/kernel_holdout_emission_blurring_pathtermination_ao.h
 	split/kernel_indirect_background.h
 	split/kernel_indirect_subsurface.h
@@ -450,6 +456,7 @@ add_library(cycles_kernel
 	${SRC_HEADERS}
 	${SRC_KERNELS_CPU_HEADERS}
 	${SRC_KERNELS_CUDA_HEADERS}
+	${SRC_KERNELS_OPENCL_HEADERS}
 	${SRC_BVH_HEADERS}
 	${SRC_CLOSURE_HEADERS}
 	${SRC_FILTER_HEADERS}
@@ -490,9 +497,11 @@ delayed_install(${CMAKE_CURRENT_SOURCE_DIR} "kernels/opencl/kernel_subsurface_sc
 delayed_install(${CMAKE_CURRENT_SOURCE_DIR} "kernels/opencl/kernel_direct_lighting.cl" ${CYCLES_INSTALL_PATH}/source/kernel/kernels/opencl)
 delayed_install(${CMAKE_CURRENT_SOURCE_DIR} "kernels/opencl/kernel_shadow_blocked_ao.cl" ${CYCLES_INSTALL_PATH}/source/kernel/kernels/opencl)
 delayed_install(${CMAKE_CURRENT_SOURCE_DIR} "kernels/opencl/kernel_shadow_blocked_dl.cl" ${CYCLES_INSTALL_PATH}/source/kernel/kernels/opencl)
+delayed_install(${CMAKE_CURRENT_SOURCE_DIR} "kernels/opencl/kernel_enqueue_inactive.cl" ${CYCLES_INSTALL_PATH}/source/kernel/kernels/opencl)
 delayed_install(${CMAKE_CURRENT_SOURCE_DIR} "kernels/opencl/kernel_next_iteration_setup.cl" ${CYCLES_INSTALL_PATH}/source/kernel/kernels/opencl)
 delayed_install(${CMAKE_CURRENT_SOURCE_DIR} "kernels/opencl/kernel_indirect_subsurface.cl" ${CYCLES_INSTALL_PATH}/source/kernel/kernels/opencl)
 delayed_install(${CMAKE_CURRENT_SOURCE_DIR} "kernels/opencl/kernel_buffer_update.cl" ${CYCLES_INSTALL_PATH}/source/kernel/kernels/opencl)
+delayed_install(${CMAKE_CURRENT_SOURCE_DIR} "kernels/opencl/kernel_split_function.h" ${CYCLES_INSTALL_PATH}/source/kernel/kernels/opencl)
 delayed_install(${CMAKE_CURRENT_SOURCE_DIR} "kernels/opencl/filter.cl" ${CYCLES_INSTALL_PATH}/source/kernel/kernels/opencl)
 delayed_install(${CMAKE_CURRENT_SOURCE_DIR} "kernels/cuda/kernel.cu" ${CYCLES_INSTALL_PATH}/source/kernel/kernels/cuda)
 delayed_install(${CMAKE_CURRENT_SOURCE_DIR} "kernels/cuda/kernel_split.cu" ${CYCLES_INSTALL_PATH}/source/kernel/kernels/cuda)

intern/cycles/kernel/closure/bsdf.h

@@ -423,6 +423,11 @@ ccl_device bool bsdf_merge(ShaderClosure *a, ShaderClosure *b)
 		case CLOSURE_BSDF_HAIR_REFLECTION_ID:
 		case CLOSURE_BSDF_HAIR_TRANSMISSION_ID:
 			return bsdf_hair_merge(a, b);
+#ifdef __PRINCIPLED__
+		case CLOSURE_BSDF_PRINCIPLED_DIFFUSE_ID:
+		case CLOSURE_BSDF_BSSRDF_PRINCIPLED_ID:
+			return bsdf_principled_diffuse_merge(a, b);
+#endif
 #ifdef __VOLUME__
 		case CLOSURE_VOLUME_HENYEY_GREENSTEIN_ID:
 			return volume_henyey_greenstein_merge(a, b);

intern/cycles/kernel/closure/bsdf_microfacet.h

@@ -288,12 +288,16 @@ ccl_device int bsdf_microfacet_ggx_setup(MicrofacetBsdf *bsdf)
 	return SD_BSDF|SD_BSDF_HAS_EVAL;
 }
 
-ccl_device int bsdf_microfacet_ggx_fresnel_setup(MicrofacetBsdf *bsdf)
+ccl_device int bsdf_microfacet_ggx_fresnel_setup(MicrofacetBsdf *bsdf, const ShaderData *sd)
 {
 	bsdf->extra->cspec0.x = saturate(bsdf->extra->cspec0.x);
 	bsdf->extra->cspec0.y = saturate(bsdf->extra->cspec0.y);
 	bsdf->extra->cspec0.z = saturate(bsdf->extra->cspec0.z);
 
+	float F0 = fresnel_dielectric_cos(1.0f, bsdf->ior);
+	float F = average(interpolate_fresnel_color(sd->I, bsdf->N, bsdf->ior, F0, bsdf->extra->cspec0));
+	bsdf->sample_weight *= F;
+
 	bsdf->alpha_x = saturate(bsdf->alpha_x);
 	bsdf->alpha_y = bsdf->alpha_x;
 
@@ -302,12 +306,16 @@ ccl_device int bsdf_microfacet_ggx_fresnel_setup(MicrofacetBsdf *bsdf)
 	return SD_BSDF|SD_BSDF_HAS_EVAL;
 }
 
-ccl_device int bsdf_microfacet_ggx_clearcoat_setup(MicrofacetBsdf *bsdf)
+ccl_device int bsdf_microfacet_ggx_clearcoat_setup(MicrofacetBsdf *bsdf, const ShaderData *sd)
 {
 	bsdf->extra->cspec0.x = saturate(bsdf->extra->cspec0.x);
 	bsdf->extra->cspec0.y = saturate(bsdf->extra->cspec0.y);
 	bsdf->extra->cspec0.z = saturate(bsdf->extra->cspec0.z);
 
+	float F0 = fresnel_dielectric_cos(1.0f, bsdf->ior);
+	float F = average(interpolate_fresnel_color(sd->I, bsdf->N, bsdf->ior, F0, bsdf->extra->cspec0));
+	bsdf->sample_weight *= 0.25f * bsdf->extra->clearcoat * F;
+
 	bsdf->alpha_x = saturate(bsdf->alpha_x);
 	bsdf->alpha_y = bsdf->alpha_x;
 
@@ -343,12 +351,16 @@ ccl_device int bsdf_microfacet_ggx_aniso_setup(MicrofacetBsdf *bsdf)
 	return SD_BSDF|SD_BSDF_HAS_EVAL;
 }
 
-ccl_device int bsdf_microfacet_ggx_aniso_fresnel_setup(MicrofacetBsdf *bsdf)
+ccl_device int bsdf_microfacet_ggx_aniso_fresnel_setup(MicrofacetBsdf *bsdf, const ShaderData *sd)
 {
 	bsdf->extra->cspec0.x = saturate(bsdf->extra->cspec0.x);
 	bsdf->extra->cspec0.y = saturate(bsdf->extra->cspec0.y);
 	bsdf->extra->cspec0.z = saturate(bsdf->extra->cspec0.z);
 
+	float F0 = fresnel_dielectric_cos(1.0f, bsdf->ior);
+	float F = average(interpolate_fresnel_color(sd->I, bsdf->N, bsdf->ior, F0, bsdf->extra->cspec0));
+	bsdf->sample_weight *= F;
+
 	bsdf->alpha_x = saturate(bsdf->alpha_x);
 	bsdf->alpha_y = saturate(bsdf->alpha_y);
 

intern/cycles/kernel/closure/bsdf_microfacet_multi.h

@@ -40,20 +40,20 @@ ccl_device_forceinline float D_ggx_aniso(const float3 wm, const float2 alpha)
 }
 
 /* Sample slope distribution (based on page 14 of the supplemental implementation). */
-ccl_device_forceinline float2 mf_sampleP22_11(const float cosI, const float2 randU)
+ccl_device_forceinline float2 mf_sampleP22_11(const float cosI, const float randx, const float randy)
 {
 	if(cosI > 0.9999f || fabsf(cosI) < 1e-6f) {
-		const float r = sqrtf(randU.x / max(1.0f - randU.x, 1e-7f));
-		const float phi = M_2PI_F * randU.y;
+		const float r = sqrtf(randx / max(1.0f - randx, 1e-7f));
+		const float phi = M_2PI_F * randy;
 		return make_float2(r*cosf(phi), r*sinf(phi));
 	}
 
-	const float sinI = sqrtf(1.0f - cosI*cosI);
+	const float sinI = safe_sqrtf(1.0f - cosI*cosI);
 	const float tanI = sinI/cosI;
 	const float projA = 0.5f * (cosI + 1.0f);
 	if(projA < 0.0001f)
 		return make_float2(0.0f, 0.0f);
-	const float A = 2.0f*randU.x*projA / cosI - 1.0f;
+	const float A = 2.0f*randx*projA / cosI - 1.0f;
 	float tmp = A*A-1.0f;
 	if(fabsf(tmp) < 1e-7f)
 		return make_float2(0.0f, 0.0f);
@@ -64,24 +64,24 @@ ccl_device_forceinline float2 mf_sampleP22_11(const float cosI, const float2 ran
 	const float slopeX = (A < 0.0f || slopeX2 > 1.0f/tanI)? (tanI*tmp - D) : slopeX2;
 
 	float U2;
-	if(randU.y >= 0.5f)
-		U2 = 2.0f*(randU.y - 0.5f);
+	if(randy >= 0.5f)
+		U2 = 2.0f*(randy - 0.5f);
 	else
-		U2 = 2.0f*(0.5f - randU.y);
+		U2 = 2.0f*(0.5f - randy);
 	const float z = (U2*(U2*(U2*0.27385f-0.73369f)+0.46341f)) / (U2*(U2*(U2*0.093073f+0.309420f)-1.0f)+0.597999f);
 	const float slopeY = z * sqrtf(1.0f + slopeX*slopeX);
 
-	if(randU.y >= 0.5f)
+	if(randy >= 0.5f)
 		return make_float2(slopeX, slopeY);
 	else
 		return make_float2(slopeX, -slopeY);
 }
 
 /* Visible normal sampling for the GGX distribution (based on page 7 of the supplemental implementation). */
-ccl_device_forceinline float3 mf_sample_vndf(const float3 wi, const float2 alpha, const float2 randU)
+ccl_device_forceinline float3 mf_sample_vndf(const float3 wi, const float2 alpha, const float randx, const float randy)
 {
 	const float3 wi_11 = normalize(make_float3(alpha.x*wi.x, alpha.y*wi.y, wi.z));
-	const float2 slope_11 = mf_sampleP22_11(wi_11.z, randU);
+	const float2 slope_11 = mf_sampleP22_11(wi_11.z, randx, randy);
 
 	const float3 cossin_phi = safe_normalize(make_float3(wi_11.x, wi_11.y, 0.0f));
 	const float slope_x = alpha.x*(cossin_phi.x * slope_11.x - cossin_phi.y * slope_11.y);
@@ -91,18 +91,15 @@ ccl_device_forceinline float3 mf_sample_vndf(const float3 wi, const float2 alpha
 	return normalize(make_float3(-slope_x, -slope_y, 1.0f));
 }
 
-/* === Phase functions: Glossy, Diffuse and Glass === */
+/* === Phase functions: Glossy and Glass === */
 
-/* Phase function for reflective materials, either without a fresnel term (for compatibility) or with the conductive fresnel term. */
-ccl_device_forceinline float3 mf_sample_phase_glossy(const float3 wi, float3 *n, float3 *k, float3 *weight, const float3 wm)
+/* Phase function for reflective materials. */
+ccl_device_forceinline float3 mf_sample_phase_glossy(const float3 wi, float3 *weight, const float3 wm)
 {
-	if(n && k)
-		*weight *= fresnel_conductor(dot(wi, wm), *n, *k);
-
 	return -wi + 2.0f * wm * dot(wi, wm);
 }
 
-ccl_device_forceinline float3 mf_eval_phase_glossy(const float3 w, const float lambda, const float3 wo, const float2 alpha, float3 *n, float3 *k)
+ccl_device_forceinline float3 mf_eval_phase_glossy(const float3 w, const float lambda, const float3 wo, const float2 alpha)
 {
 	if(w.z > 0.9999f)
 		return make_float3(0.0f, 0.0f, 0.0f);
@@ -123,30 +120,9 @@ ccl_device_forceinline float3 mf_eval_phase_glossy(const float3 w, const float l
 	else
 		phase *= D_ggx_aniso(wh, alpha);
 
-	if(n && k) {
-		/* Apply conductive fresnel term. */
-		return phase * fresnel_conductor(dotW_WH, *n, *k);
-	}
-
 	return make_float3(phase, phase, phase);
 }
 
-/* Phase function for rough lambertian diffuse surfaces. */
-ccl_device_forceinline float3 mf_sample_phase_diffuse(const float3 wm, const float randu, const float randv)
-{
-	float3 tm, bm;
-	make_orthonormals(wm, &tm, &bm);
-
-	float2 disk = concentric_sample_disk(randu, randv);
-	return disk.x*tm + disk.y*bm + safe_sqrtf(1.0f - disk.x*disk.x - disk.y*disk.y)*wm;
-}
-
-ccl_device_forceinline float3 mf_eval_phase_diffuse(const float3 w, const float3 wm)
-{
-	const float v = max(0.0f, dot(w, wm)) * M_1_PI_F;
-	return make_float3(v, v, v);
-}
-
 /* Phase function for dielectric transmissive materials, including both reflection and refraction according to the dielectric fresnel term. */
 ccl_device_forceinline float3 mf_sample_phase_glass(const float3 wi, const float eta, const float3 wm, const float randV, bool *outside)
 {
@@ -269,40 +245,69 @@ ccl_device_forceinline float mf_ggx_albedo(float r)
 	return saturate(albedo);
 }
 
+ccl_device_inline float mf_ggx_transmission_albedo(float a, float ior)
+{
+	if(ior < 1.0f) {
+		ior = 1.0f/ior;
+	}
+	a = saturate(a);
+	ior = clamp(ior, 1.0f, 3.0f);
+	float I_1 = 0.0476898f*expf(-0.978352f*(ior-0.65657f)*(ior-0.65657f)) - 0.033756f*ior + 0.993261f;
+	float R_1 = (((0.116991f*a - 0.270369f)*a + 0.0501366f)*a - 0.00411511f)*a + 1.00008f;
+	float I_2 = (((-2.08704f*ior + 26.3298f)*ior - 127.906f)*ior + 292.958f)*ior - 287.946f + 199.803f/(ior*ior) - 101.668f/(ior*ior*ior);
+	float R_2 = ((((5.3725f*a -24.9307f)*a + 22.7437f)*a - 3.40751f)*a + 0.0986325f)*a + 0.00493504f;
+
+	return saturate(1.0f + I_2*R_2*0.0019127f - (1.0f - I_1)*(1.0f - R_1)*9.3205f);
+}
+
 ccl_device_forceinline float mf_ggx_pdf(const float3 wi, const float3 wo, const float alpha)
 {
 	float D = D_ggx(normalize(wi+wo), alpha);
 	float lambda = mf_lambda(wi, make_float2(alpha, alpha));
+	float singlescatter = 0.25f * D / max((1.0f + lambda) * wi.z, 1e-7f);
+
+	float multiscatter = wo.z * M_1_PI_F;
+
 	float albedo = mf_ggx_albedo(alpha);
-	return 0.25f * D / max((1.0f + lambda) * wi.z, 1e-7f) + (1.0f - albedo) * wo.z;
+	return albedo*singlescatter + (1.0f - albedo)*multiscatter;
 }
 
 ccl_device_forceinline float mf_ggx_aniso_pdf(const float3 wi, const float3 wo, const float2 alpha)
 {
-	return 0.25f * D_ggx_aniso(normalize(wi+wo), alpha) / ((1.0f + mf_lambda(wi, alpha)) * wi.z) + (1.0f - mf_ggx_albedo(sqrtf(alpha.x*alpha.y))) * wo.z;
-}
+	float D = D_ggx_aniso(normalize(wi+wo), alpha);
+	float lambda = mf_lambda(wi, alpha);
+	float singlescatter = 0.25f * D / max((1.0f + lambda) * wi.z, 1e-7f);
 
-ccl_device_forceinline float mf_diffuse_pdf(const float3 wo)
-{
-	return M_1_PI_F * wo.z;
+	float multiscatter = wo.z * M_1_PI_F;
+
+	float albedo = mf_ggx_albedo(sqrtf(alpha.x*alpha.y));
+	return albedo*singlescatter + (1.0f - albedo)*multiscatter;
 }
 
 ccl_device_forceinline float mf_glass_pdf(const float3 wi, const float3 wo, const float alpha, const float eta)
 {
-	float3 wh;
-	float fresnel;
-	if(wi.z*wo.z > 0.0f) {
-		wh = normalize(wi + wo);
-		fresnel = fresnel_dielectric_cos(dot(wi, wh), eta);
-	}
-	else {
-		wh = normalize(wi + wo*eta);
-		fresnel = 1.0f - fresnel_dielectric_cos(dot(wi, wh), eta);
-	}
+	bool reflective = (wi.z*wo.z > 0.0f);
+
+	float wh_len;
+	float3 wh = normalize_len(wi + (reflective? wo : (wo*eta)), &wh_len);
 	if(wh.z < 0.0f)
 		wh = -wh;
 	float3 r_wi = (wi.z < 0.0f)? -wi: wi;
-	return fresnel * max(0.0f, dot(r_wi, wh)) * D_ggx(wh, alpha) / ((1.0f + mf_lambda(r_wi, make_float2(alpha, alpha))) * r_wi.z) + fabsf(wo.z);
+	float lambda = mf_lambda(r_wi, make_float2(alpha, alpha));
+	float D = D_ggx(wh, alpha);
+	float fresnel = fresnel_dielectric_cos(dot(r_wi, wh), eta);
+
+	float multiscatter = fabsf(wo.z * M_1_PI_F);
+	if(reflective) {
+		float singlescatter = 0.25f * D / max((1.0f + lambda) * r_wi.z, 1e-7f);
+		float albedo = mf_ggx_albedo(alpha);
+		return fresnel * (albedo*singlescatter + (1.0f - albedo)*multiscatter);
+	}
+	else {
+		float singlescatter = fabsf(dot(r_wi, wh)*dot(wo, wh) * D * eta*eta / max((1.0f + lambda) * r_wi.z * wh_len*wh_len, 1e-7f));
+		float albedo = mf_ggx_transmission_albedo(alpha, eta);
+		return (1.0f - fresnel) * (albedo*singlescatter + (1.0f - albedo)*multiscatter);
+	}
 }
 
 /* === Actual random walk implementations, one version of mf_eval and mf_sample per phase function. === */
@@ -315,13 +320,6 @@ ccl_device_forceinline float mf_glass_pdf(const float3 wi, const float3 wo, cons
 #define MF_MULTI_GLASS
 #include "kernel/closure/bsdf_microfacet_multi_impl.h"
 
-/* The diffuse phase function is not implemented as a node yet. */
-#if 0
-#define MF_PHASE_FUNCTION diffuse
-#define MF_MULTI_DIFFUSE
-#include "kernel/closure/bsdf_microfacet_multi_impl.h"
-#endif
-
 #define MF_PHASE_FUNCTION glossy
 #define MF_MULTI_GLOSSY
 #include "kernel/closure/bsdf_microfacet_multi_impl.h"
@@ -362,13 +360,17 @@ ccl_device int bsdf_microfacet_multi_ggx_aniso_setup(MicrofacetBsdf *bsdf)
 	return bsdf_microfacet_multi_ggx_common_setup(bsdf);
 }
 
-ccl_device int bsdf_microfacet_multi_ggx_aniso_fresnel_setup(MicrofacetBsdf *bsdf)
+ccl_device int bsdf_microfacet_multi_ggx_aniso_fresnel_setup(MicrofacetBsdf *bsdf, const ShaderData *sd)
 {
 	if(is_zero(bsdf->T))
 		bsdf->T = make_float3(1.0f, 0.0f, 0.0f);
 
 	bsdf->type = CLOSURE_BSDF_MICROFACET_MULTI_GGX_FRESNEL_ID;
 
+	float F0 = fresnel_dielectric_cos(1.0f, bsdf->ior);
+	float F = average(interpolate_fresnel_color(sd->I, bsdf->N, bsdf->ior, F0, bsdf->extra->cspec0));
+	bsdf->sample_weight *= F;
+
 	return bsdf_microfacet_multi_ggx_common_setup(bsdf);
 }
 
@@ -381,12 +383,16 @@ ccl_device int bsdf_microfacet_multi_ggx_setup(MicrofacetBsdf *bsdf)
 	return bsdf_microfacet_multi_ggx_common_setup(bsdf);
 }
 
-ccl_device int bsdf_microfacet_multi_ggx_fresnel_setup(MicrofacetBsdf *bsdf)
+ccl_device int bsdf_microfacet_multi_ggx_fresnel_setup(MicrofacetBsdf *bsdf, const ShaderData *sd)
 {
 	bsdf->alpha_y = bsdf->alpha_x;
 
 	bsdf->type = CLOSURE_BSDF_MICROFACET_MULTI_GGX_FRESNEL_ID;
 
+	float F0 = fresnel_dielectric_cos(1.0f, bsdf->ior);
+	float F = average(interpolate_fresnel_color(sd->I, bsdf->N, bsdf->ior, F0, bsdf->extra->cspec0));
+	bsdf->sample_weight *= F;
+
 	return bsdf_microfacet_multi_ggx_common_setup(bsdf);
 }
 
@@ -428,7 +434,7 @@ ccl_device float3 bsdf_microfacet_multi_ggx_eval_reflect(const ShaderClosure *sc
 		*pdf = mf_ggx_aniso_pdf(localI, localO, make_float2(bsdf->alpha_x, bsdf->alpha_y));
 	else
 		*pdf = mf_ggx_pdf(localI, localO, bsdf->alpha_x);
-	return mf_eval_glossy(localI, localO, true, bsdf->extra->color, bsdf->alpha_x, bsdf->alpha_y, lcg_state, NULL, NULL, bsdf->ior, use_fresnel, bsdf->extra->cspec0);
+	return mf_eval_glossy(localI, localO, true, bsdf->extra->color, bsdf->alpha_x, bsdf->alpha_y, lcg_state, bsdf->ior, use_fresnel, bsdf->extra->cspec0);
 }
 
 ccl_device int bsdf_microfacet_multi_ggx_sample(KernelGlobals *kg, const ShaderClosure *sc, float3 Ng, float3 I, float3 dIdx, float3 dIdy, float randu, float randv, float3 *eval, float3 *omega_in, float3 *domega_in_dx, float3 *domega_in_dy, float *pdf, ccl_addr_space uint *lcg_state)
@@ -442,6 +448,10 @@ ccl_device int bsdf_microfacet_multi_ggx_sample(KernelGlobals *kg, const ShaderC
 		*omega_in = 2*dot(Z, I)*Z - I;
 		*pdf = 1e6f;
 		*eval = make_float3(1e6f, 1e6f, 1e6f);
+#ifdef __RAY_DIFFERENTIALS__
+		*domega_in_dx = (2 * dot(Z, dIdx)) * Z - dIdx;
+		*domega_in_dy = (2 * dot(Z, dIdy)) * Z - dIdy;
+#endif
 		return LABEL_REFLECT|LABEL_SINGULAR;
 	}
 
@@ -456,7 +466,7 @@ ccl_device int bsdf_microfacet_multi_ggx_sample(KernelGlobals *kg, const ShaderC
 	float3 localI = make_float3(dot(I, X), dot(I, Y), dot(I, Z));
 	float3 localO;
 
-	*eval = mf_sample_glossy(localI, &localO, bsdf->extra->color, bsdf->alpha_x, bsdf->alpha_y, lcg_state, NULL, NULL, bsdf->ior, use_fresnel, bsdf->extra->cspec0);
+	*eval = mf_sample_glossy(localI, &localO, bsdf->extra->color, bsdf->alpha_x, bsdf->alpha_y, lcg_state, bsdf->ior, use_fresnel, bsdf->extra->cspec0);
 	if(is_aniso)
 		*pdf = mf_ggx_aniso_pdf(localI, localO, make_float2(bsdf->alpha_x, bsdf->alpha_y));
 	else
@@ -464,6 +474,7 @@ ccl_device int bsdf_microfacet_multi_ggx_sample(KernelGlobals *kg, const ShaderC
 	*eval *= *pdf;
 
 	*omega_in = X*localO.x + Y*localO.y + Z*localO.z;
+
 #ifdef __RAY_DIFFERENTIALS__
 	*domega_in_dx = (2 * dot(Z, dIdx)) * Z - dIdx;
 	*domega_in_dy = (2 * dot(Z, dIdy)) * Z - dIdy;
@@ -487,7 +498,7 @@ ccl_device int bsdf_microfacet_multi_ggx_glass_setup(MicrofacetBsdf *bsdf)
 	return SD_BSDF|SD_BSDF_HAS_EVAL|SD_BSDF_NEEDS_LCG;
 }
 
-ccl_device int bsdf_microfacet_multi_ggx_glass_fresnel_setup(MicrofacetBsdf *bsdf)
+ccl_device int bsdf_microfacet_multi_ggx_glass_fresnel_setup(MicrofacetBsdf *bsdf, const ShaderData *sd)
 {
 	bsdf->alpha_x = clamp(bsdf->alpha_x, 1e-4f, 1.0f);
 	bsdf->alpha_y = bsdf->alpha_x;
@@ -501,6 +512,10 @@ ccl_device int bsdf_microfacet_multi_ggx_glass_fresnel_setup(MicrofacetBsdf *bsd
 
 	bsdf->type = CLOSURE_BSDF_MICROFACET_MULTI_GGX_GLASS_FRESNEL_ID;
 
+	float F0 = fresnel_dielectric_cos(1.0f, bsdf->ior);
+	float F = average(interpolate_fresnel_color(sd->I, bsdf->N, bsdf->ior, F0, bsdf->extra->cspec0));
+	bsdf->sample_weight *= F;
+
 	return SD_BSDF|SD_BSDF_HAS_EVAL|SD_BSDF_NEEDS_LCG;
 }
 

intern/cycles/kernel/closure/bsdf_microfacet_multi_impl.h

@@ -26,24 +26,16 @@
  * the balance heuristic isn't necessarily optimal anymore.
  */
 ccl_device_forceinline float3 MF_FUNCTION_FULL_NAME(mf_eval)(
-        float3 wi,
-        float3 wo,
-        const bool wo_outside,
-        const float3 color,
-        const float alpha_x,
-        const float alpha_y,
-         ccl_addr_space uint *lcg_state
-#ifdef MF_MULTI_GLASS
-        , const float eta
-        , bool use_fresnel
-        , const float3 cspec0
-#elif defined(MF_MULTI_GLOSSY)
-         , float3 *n, float3 *k
-         , const float eta
-         , bool use_fresnel
-         , const float3 cspec0
-#endif
-)
+	float3 wi,
+	float3 wo,
+	const bool wo_outside,
+	const float3 color,
+	const float alpha_x,
+	const float alpha_y,
+	ccl_addr_space uint *lcg_state,
+	const float eta,
+	bool use_fresnel,
+	const float3 cspec0)
 {
 	/* Evaluating for a shallower incoming direction produces less noise, and the properties of the BSDF guarantee reciprocity. */
 	bool swapped = false;
@@ -77,44 +69,29 @@ ccl_device_forceinline float3 MF_FUNCTION_FULL_NAME(mf_eval)(
 	/* Analytically compute single scattering for lower noise. */
 	float3 eval;
 	float3 throughput = make_float3(1.0f, 1.0f, 1.0f);
+	const float3 wh = normalize(wi+wo);
 #ifdef MF_MULTI_GLASS
 	eval = mf_eval_phase_glass(-wi, lambda_r, wo, wo_outside, alpha, eta);
 	if(wo_outside)
 		eval *= -lambda_r / (shadowing_lambda - lambda_r);
 	else
 		eval *= -lambda_r * beta(-lambda_r, shadowing_lambda+1.0f);
-
-	float F0 = fresnel_dielectric_cos(1.0f, eta);
-	if(use_fresnel) {
-		throughput = interpolate_fresnel_color(wi, normalize(wi + wo), eta, F0, cspec0);
-
-		eval *= throughput;
-	}
-#elif defined(MF_MULTI_DIFFUSE)
-	/* Diffuse has no special closed form for the single scattering bounce */
-	eval = make_float3(0.0f, 0.0f, 0.0f);
 #else /* MF_MULTI_GLOSSY */
-	const float3 wh = normalize(wi+wo);
 	const float G2 = 1.0f / (1.0f - (lambda_r + 1.0f) + shadowing_lambda);
 	float val = G2 * 0.25f / wi.z;
 	if(alpha.x == alpha.y)
 		val *= D_ggx(wh, alpha.x);
 	else
 		val *= D_ggx_aniso(wh, alpha);
-	if(n && k) {
-		eval = fresnel_conductor(dot(wh, wi), *n, *k) * val;
-	}
-	else {
-		eval = make_float3(val, val, val);
-	}
+	eval = make_float3(val, val, val);
+#endif
 
 	float F0 = fresnel_dielectric_cos(1.0f, eta);
 	if(use_fresnel) {
 		throughput = interpolate_fresnel_color(wi, wh, eta, F0, cspec0);
 
-		eval = throughput * val;
+		eval *= throughput;
 	}
-#endif
 
 	float3 wr = -wi;
 	float hr = 1.0f;
@@ -123,19 +100,15 @@ ccl_device_forceinline float3 MF_FUNCTION_FULL_NAME(mf_eval)(
 	bool outside = true;
 
 	for(int order = 0; order < 10; order++) {
-		/* Sample microfacet height and normal */
-		if(!mf_sample_height(wr, &hr, &C1_r, &G1_r, &lambda_r, lcg_step_float_addrspace(lcg_state)))
+		/* Sample microfacet height. */
+		float height_rand = lcg_step_float_addrspace(lcg_state);
+		if(!mf_sample_height(wr, &hr, &C1_r, &G1_r, &lambda_r, height_rand))
 			break;
-		float3 wm = mf_sample_vndf(-wr, alpha, make_float2(lcg_step_float_addrspace(lcg_state),
-		                                                   lcg_step_float_addrspace(lcg_state)));
+		/* Sample microfacet normal. */
+		float vndf_rand_y = lcg_step_float_addrspace(lcg_state);
+		float vndf_rand_x = lcg_step_float_addrspace(lcg_state);
+		float3 wm = mf_sample_vndf(-wr, alpha, vndf_rand_x, vndf_rand_y);
 
-#ifdef MF_MULTI_DIFFUSE
-		if(order == 0) {
-			/* Compute single-scattering for diffuse. */
-			const float G2_G1 = -lambda_r / (shadowing_lambda - lambda_r);
-			eval += throughput * G2_G1 * mf_eval_phase_diffuse(wo, wm);
-		}
-#endif
 #ifdef MF_MULTI_GLASS
 		if(order == 0 && use_fresnel) {
 			/* Evaluate amount of scattering towards wo on this microfacet. */
@@ -156,10 +129,8 @@ ccl_device_forceinline float3 MF_FUNCTION_FULL_NAME(mf_eval)(
 				phase = mf_eval_phase_glass(wr, lambda_r,  wo,  wo_outside, alpha, eta);
 			else
 				phase = mf_eval_phase_glass(wr, lambda_r, -wo, !wo_outside, alpha, 1.0f/eta);
-#elif defined(MF_MULTI_DIFFUSE)
-			phase = mf_eval_phase_diffuse(wo, wm);
 #else /* MF_MULTI_GLOSSY */
-			phase = mf_eval_phase_glossy(wr, lambda_r, wo, alpha, n, k) * throughput;
+			phase = mf_eval_phase_glossy(wr, lambda_r, wo, alpha) * throughput;
 #endif
 			eval += throughput * phase * mf_G1(wo_outside? wo: -wo, mf_C1((outside == wo_outside)? hr: -hr), shadowing_lambda);
 		}
@@ -168,7 +139,8 @@ ccl_device_forceinline float3 MF_FUNCTION_FULL_NAME(mf_eval)(
 #ifdef MF_MULTI_GLASS
 			bool next_outside;
 			float3 wi_prev = -wr;
-			wr = mf_sample_phase_glass(-wr, outside? eta: 1.0f/eta, wm, lcg_step_float_addrspace(lcg_state), &next_outside);
+			float phase_rand = lcg_step_float_addrspace(lcg_state);
+			wr = mf_sample_phase_glass(-wr, outside? eta: 1.0f/eta, wm, phase_rand, &next_outside);
 			if(!next_outside) {
 				outside = !outside;
 				wr = -wr;
@@ -181,25 +153,17 @@ ccl_device_forceinline float3 MF_FUNCTION_FULL_NAME(mf_eval)(
 			else if(use_fresnel && order > 0) {
 				throughput *= interpolate_fresnel_color(wi_prev, wm, eta, F0, cspec0);
 			}
-#elif defined(MF_MULTI_DIFFUSE)
-			wr = mf_sample_phase_diffuse(wm,
-			                             lcg_step_float_addrspace(lcg_state),
-			                             lcg_step_float_addrspace(lcg_state));
 #else /* MF_MULTI_GLOSSY */
 			if(use_fresnel && order > 0) {
 				throughput *= interpolate_fresnel_color(-wr, wm, eta, F0, cspec0);
 			}
-			wr = mf_sample_phase_glossy(-wr, n, k, &throughput, wm);
+			wr = mf_sample_phase_glossy(-wr, &throughput, wm);
 #endif
 
 			lambda_r = mf_lambda(wr, alpha);
 
-#if defined(MF_MULTI_GLOSSY) || defined(MF_MULTI_GLASS)
 			if(!use_fresnel)
 				throughput *= color;
-#else
-			throughput *= color;
-#endif
 
 			C1_r = mf_C1(hr);
 			G1_r = mf_G1(wr, C1_r, lambda_r);
@@ -215,18 +179,16 @@ ccl_device_forceinline float3 MF_FUNCTION_FULL_NAME(mf_eval)(
  * escaped the surface in wo. The function returns the throughput between wi and wo.
  * Without reflection losses due to coloring or fresnel absorption in conductors, the sampling is optimal.
  */
-ccl_device_forceinline float3 MF_FUNCTION_FULL_NAME(mf_sample)(float3 wi, float3 *wo, const float3 color, const float alpha_x, const float alpha_y, ccl_addr_space uint *lcg_state
-#ifdef MF_MULTI_GLASS
-	, const float eta
-	, bool use_fresnel
-	, const float3 cspec0
-#elif defined(MF_MULTI_GLOSSY)
-	, float3 *n, float3 *k
-	, const float eta
-	, bool use_fresnel
-	, const float3 cspec0
-#endif
-)
+ccl_device_forceinline float3 MF_FUNCTION_FULL_NAME(mf_sample)(
+	float3 wi,
+	float3 *wo,
+	const float3 color,
+	const float alpha_x,
+	const float alpha_y,
+	ccl_addr_space uint *lcg_state,
+	const float eta,
+	bool use_fresnel,
+	const float3 cspec0)
 {
 	const float2 alpha = make_float2(alpha_x, alpha_y);
 
@@ -237,44 +199,36 @@ ccl_device_forceinline float3 MF_FUNCTION_FULL_NAME(mf_sample)(float3 wi, float3
 	float C1_r = 1.0f;
 	float G1_r = 0.0f;
 	bool outside = true;
-#ifdef MF_MULTI_GLASS
+
 	float F0 = fresnel_dielectric_cos(1.0f, eta);
 	if(use_fresnel) {
 		throughput = interpolate_fresnel_color(wi, normalize(wi + wr), eta, F0, cspec0);
 	}
-#elif defined(MF_MULTI_GLOSSY)
-	float F0 = fresnel_dielectric_cos(1.0f, eta);
-	if(use_fresnel) {
-		throughput = interpolate_fresnel_color(wi, normalize(wi + wr), eta, F0, cspec0);
-	}
-#endif
 
 	int order;
 	for(order = 0; order < 10; order++) {
 		/* Sample microfacet height. */
-		if(!mf_sample_height(wr, &hr, &C1_r, &G1_r, &lambda_r, lcg_step_float_addrspace(lcg_state))) {
+		float height_rand = lcg_step_float_addrspace(lcg_state);
+		if(!mf_sample_height(wr, &hr, &C1_r, &G1_r, &lambda_r, height_rand)) {
 			/* The random walk has left the surface. */
 			*wo = outside? wr: -wr;
 			return throughput;
 		}
 		/* Sample microfacet normal. */
-		float3 wm = mf_sample_vndf(-wr, alpha, make_float2(lcg_step_float_addrspace(lcg_state),
-		                                                   lcg_step_float_addrspace(lcg_state)));
+		float vndf_rand_y = lcg_step_float_addrspace(lcg_state);
+		float vndf_rand_x = lcg_step_float_addrspace(lcg_state);
+		float3 wm = mf_sample_vndf(-wr, alpha, vndf_rand_x, vndf_rand_y);
 
 		/* First-bounce color is already accounted for in mix weight. */
-#if defined(MF_MULTI_GLASS) || defined(MF_MULTI_GLOSSY)
 		if(!use_fresnel && order > 0)
 			throughput *= color;
-#else
-		if(order > 0)
-			throughput *= color;
-#endif
 
 		/* Bounce from the microfacet. */
 #ifdef MF_MULTI_GLASS
 		bool next_outside;
 		float3 wi_prev = -wr;
-		wr = mf_sample_phase_glass(-wr, outside? eta: 1.0f/eta, wm, lcg_step_float_addrspace(lcg_state), &next_outside);
+		float phase_rand = lcg_step_float_addrspace(lcg_state);
+		wr = mf_sample_phase_glass(-wr, outside? eta: 1.0f/eta, wm, phase_rand, &next_outside);
 		if(!next_outside) {
 			hr = -hr;
 			wr = -wr;
@@ -294,10 +248,6 @@ ccl_device_forceinline float3 MF_FUNCTION_FULL_NAME(mf_sample)(float3 wi, float3
 					throughput *= t_color;
 			}
 		}
-#elif defined(MF_MULTI_DIFFUSE)
-		wr = mf_sample_phase_diffuse(wm,
-		                             lcg_step_float_addrspace(lcg_state),
-		                             lcg_step_float_addrspace(lcg_state));
 #else /* MF_MULTI_GLOSSY */
 		if(use_fresnel) {
 			float3 t_color = interpolate_fresnel_color(-wr, wm, eta, F0, cspec0);
@@ -307,7 +257,7 @@ ccl_device_forceinline float3 MF_FUNCTION_FULL_NAME(mf_sample)(float3 wi, float3
 			else
 				throughput *= t_color;
 		}
-		wr = mf_sample_phase_glossy(-wr, n, k, &throughput, wm);
+		wr = mf_sample_phase_glossy(-wr, &throughput, wm);
 #endif
 
 		/* Update random walk parameters. */
@@ -319,6 +269,5 @@ ccl_device_forceinline float3 MF_FUNCTION_FULL_NAME(mf_sample)(float3 wi, float3
 }
 
 #undef MF_MULTI_GLASS
-#undef MF_MULTI_DIFFUSE
 #undef MF_MULTI_GLOSSY
 #undef MF_PHASE_FUNCTION

intern/cycles/kernel/closure/bsdf_principled_diffuse.h

@@ -58,6 +58,14 @@ ccl_device int bsdf_principled_diffuse_setup(PrincipledDiffuseBsdf *bsdf)
 	return SD_BSDF|SD_BSDF_HAS_EVAL;
 }
 
+ccl_device bool bsdf_principled_diffuse_merge(const ShaderClosure *a, const ShaderClosure *b)
+{
+	const PrincipledDiffuseBsdf *bsdf_a = (const PrincipledDiffuseBsdf*)a;
+	const PrincipledDiffuseBsdf *bsdf_b = (const PrincipledDiffuseBsdf*)b;
+
+	return (isequal_float3(bsdf_a->N, bsdf_b->N) && bsdf_a->roughness == bsdf_b->roughness);
+}
+
 ccl_device float3 bsdf_principled_diffuse_eval_reflect(const ShaderClosure *sc, const float3 I,
 	const float3 omega_in, float *pdf)
 {

intern/cycles/kernel/filter/filter_features.h

@@ -16,7 +16,7 @@
 
  CCL_NAMESPACE_BEGIN
 
-#define ccl_get_feature(buffer, pass) buffer[(pass)*pass_stride]
+#define ccl_get_feature(buffer, pass) (buffer)[(pass)*pass_stride]
 
 /* Loop over the pixels in the range [low.x, high.x) x [low.y, high.y).
  * pixel_buffer always points to the current pixel in the first pass. */
@@ -28,11 +28,15 @@
                                  pixel_buffer += buffer_w - (high.x - low.x); \
                              }
 
-ccl_device_inline void filter_get_features(int2 pixel, ccl_global float ccl_restrict_ptr buffer, float *features, float ccl_restrict_ptr mean, int pass_stride)
+ccl_device_inline void filter_get_features(int2 pixel,
+                                           const ccl_global float *ccl_restrict buffer,
+                                           float *features,
+                                           const float *ccl_restrict mean,
+                                           int pass_stride)
 {
 	features[0] = pixel.x;
 	features[1] = pixel.y;
-	features[2] = ccl_get_feature(buffer, 0);
+	features[2] = fabsf(ccl_get_feature(buffer, 0));
 	features[3] = ccl_get_feature(buffer, 1);
 	features[4] = ccl_get_feature(buffer, 2);
 	features[5] = ccl_get_feature(buffer, 3);
@@ -46,11 +50,15 @@ ccl_device_inline void filter_get_features(int2 pixel, ccl_global float ccl_rest
 	}
 }
 
-ccl_device_inline void filter_get_feature_scales(int2 pixel, ccl_global float ccl_restrict_ptr buffer, float *scales, float ccl_restrict_ptr mean, int pass_stride)
+ccl_device_inline void filter_get_feature_scales(int2 pixel,
+                                                 const ccl_global float *ccl_restrict buffer,
+                                                 float *scales,
+                                                 const float *ccl_restrict mean,
+                                                 int pass_stride)
 {
 	scales[0] = fabsf(pixel.x - mean[0]);
 	scales[1] = fabsf(pixel.y - mean[1]);
-	scales[2] = fabsf(ccl_get_feature(buffer, 0) - mean[2]);
+	scales[2] = fabsf(fabsf(ccl_get_feature(buffer, 0)) - mean[2]);
 	scales[3] = len_squared(make_float3(ccl_get_feature(buffer, 1) - mean[3],
 	                                    ccl_get_feature(buffer, 2) - mean[4],
 	                                    ccl_get_feature(buffer, 3) - mean[5]));
@@ -70,26 +78,15 @@ ccl_device_inline void filter_calculate_scale(float *scale)
 	scale[3] = scale[4] = scale[5] = 1.0f/max(sqrtf(scale[3]), 0.01f);
 }
 
-ccl_device_inline float3 filter_get_pixel_color(ccl_global float ccl_restrict_ptr buffer, int pass_stride)
+ccl_device_inline float3 filter_get_color(const ccl_global float *ccl_restrict buffer,
+                                          int pass_stride)
 {
-	return make_float3(ccl_get_feature(buffer, 0), ccl_get_feature(buffer, 1), ccl_get_feature(buffer, 2));
-}
-
-ccl_device_inline float filter_get_pixel_variance(ccl_global float ccl_restrict_ptr buffer, int pass_stride)
-{
-	return average(make_float3(ccl_get_feature(buffer, 0), ccl_get_feature(buffer, 1), ccl_get_feature(buffer, 2)));
-}
-
-ccl_device_inline bool filter_firefly_rejection(float3 pixel_color, float pixel_variance, float3 center_color, float sqrt_center_variance)
-{
-	float color_diff = average(fabs(pixel_color - center_color));
-	float variance = sqrt_center_variance + sqrtf(pixel_variance) + 0.005f;
-	return (color_diff > 3.0f*variance);
+	return make_float3(ccl_get_feature(buffer, 8), ccl_get_feature(buffer, 9), ccl_get_feature(buffer, 10));
 }
 
 ccl_device_inline void design_row_add(float *design_row,
                                       int rank,
-                                      ccl_global float ccl_restrict_ptr transform,
+                                      const ccl_global float *ccl_restrict transform,
                                       int stride,
                                       int row,
                                       float feature)
@@ -101,20 +98,20 @@ ccl_device_inline void design_row_add(float *design_row,
 
 /* Fill the design row. */
 ccl_device_inline void filter_get_design_row_transform(int2 p_pixel,
-                                                       ccl_global float ccl_restrict_ptr p_buffer,
+                                                       const ccl_global float *ccl_restrict p_buffer,
                                                        int2 q_pixel,
-                                                       ccl_global float ccl_restrict_ptr q_buffer,
+                                                       const ccl_global float *ccl_restrict q_buffer,
                                                        int pass_stride,
                                                        int rank,
                                                        float *design_row,
-                                                       ccl_global float ccl_restrict_ptr transform,
+                                                       const ccl_global float *ccl_restrict transform,
                                                        int stride)
 {
 	design_row[0] = 1.0f;
 	math_vector_zero(design_row+1, rank);
 	design_row_add(design_row, rank, transform, stride, 0, q_pixel.x - p_pixel.x);
 	design_row_add(design_row, rank, transform, stride, 1, q_pixel.y - p_pixel.y);
-	design_row_add(design_row, rank, transform, stride, 2, ccl_get_feature(q_buffer, 0) - ccl_get_feature(p_buffer, 0));
+	design_row_add(design_row, rank, transform, stride, 2, fabsf(ccl_get_feature(q_buffer, 0)) - fabsf(ccl_get_feature(p_buffer, 0)));
 	design_row_add(design_row, rank, transform, stride, 3, ccl_get_feature(q_buffer, 1) - ccl_get_feature(p_buffer, 1));
 	design_row_add(design_row, rank, transform, stride, 4, ccl_get_feature(q_buffer, 2) - ccl_get_feature(p_buffer, 2));
 	design_row_add(design_row, rank, transform, stride, 5, ccl_get_feature(q_buffer, 3) - ccl_get_feature(p_buffer, 3));

intern/cycles/kernel/filter/filter_features_sse.h

@@ -33,11 +33,16 @@ CCL_NAMESPACE_BEGIN
                                      pixel_buffer += buffer_w - (pixel.x - low.x); \
                                  }
 
-ccl_device_inline void filter_get_features_sse(__m128 x, __m128 y, __m128 active_pixels, float ccl_restrict_ptr buffer, __m128 *features, __m128 ccl_restrict_ptr mean, int pass_stride)
+ccl_device_inline void filter_get_features_sse(__m128 x, __m128 y,
+                                               __m128 active_pixels,
+                                               const float *ccl_restrict buffer,
+                                               __m128 *features,
+                                               const __m128 *ccl_restrict mean,
+                                               int pass_stride)
 {
 	features[0] = x;
 	features[1] = y;
-	features[2] = ccl_get_feature_sse(0);
+	features[2] = _mm_fabs_ps(ccl_get_feature_sse(0));
 	features[3] = ccl_get_feature_sse(1);
 	features[4] = ccl_get_feature_sse(2);
 	features[5] = ccl_get_feature_sse(3);
@@ -53,12 +58,17 @@ ccl_device_inline void filter_get_features_sse(__m128 x, __m128 y, __m128 active
 		features[i] = _mm_mask_ps(features[i], active_pixels);
 }
 
-ccl_device_inline void filter_get_feature_scales_sse(__m128 x, __m128 y, __m128 active_pixels, float ccl_restrict_ptr buffer, __m128 *scales, __m128 ccl_restrict_ptr mean, int pass_stride)
+ccl_device_inline void filter_get_feature_scales_sse(__m128 x, __m128 y,
+                                                     __m128 active_pixels,
+                                                     const float *ccl_restrict buffer,
+                                                     __m128 *scales,
+                                                     const __m128 *ccl_restrict mean,
+                                                     int pass_stride)
 {
 	scales[0] = _mm_mask_ps(_mm_fabs_ps(_mm_sub_ps(x, mean[0])), active_pixels);
 	scales[1] = _mm_mask_ps(_mm_fabs_ps(_mm_sub_ps(y, mean[1])), active_pixels);
 
-	scales[2] = _mm_mask_ps(_mm_fabs_ps(_mm_sub_ps(ccl_get_feature_sse(0), mean[2])), active_pixels);
+	scales[2] = _mm_mask_ps(_mm_fabs_ps(_mm_sub_ps(_mm_fabs_ps(ccl_get_feature_sse(0)), mean[2])), active_pixels);
 
 	__m128 diff, scale;
 	diff = _mm_sub_ps(ccl_get_feature_sse(1), mean[3]);

intern/cycles/kernel/filter/filter_kernel.h

@@ -48,22 +48,3 @@
 #else
 #  include "kernel/filter/filter_nlm_gpu.h"
 #endif
-
-CCL_NAMESPACE_BEGIN
-
-ccl_device void kernel_filter_divide_combined(int x, int y, int sample, ccl_global float *buffers, int offset, int stride, int pass_stride, int no_denoising_offset)
-{
-	ccl_global float *combined_buffer = buffers + (offset + y*stride + x);
-	float fac = sample / combined_buffer[3];
-	combined_buffer[0] *= fac;
-	combined_buffer[1] *= fac;
-	combined_buffer[2] *= fac;
-	combined_buffer[3] *= fac;
-	if(no_denoising_offset) {
-		combined_buffer[0] += combined_buffer[no_denoising_offset+0];
-		combined_buffer[1] += combined_buffer[no_denoising_offset+1];
-		combined_buffer[2] += combined_buffer[no_denoising_offset+2];
-	}
-}
-
-CCL_NAMESPACE_END

intern/cycles/kernel/filter/filter_nlm_cpu.h

@@ -16,27 +16,39 @@
 
 CCL_NAMESPACE_BEGIN
 
-ccl_device_inline void kernel_filter_nlm_calc_difference(int dx, int dy, float ccl_restrict_ptr weightImage, float ccl_restrict_ptr varianceImage, float *differenceImage, int4 rect, int w, int channel_offset, float a, float k_2)
+ccl_device_inline void kernel_filter_nlm_calc_difference(int dx, int dy,
+                                                         const float *ccl_restrict weight_image,
+                                                         const float *ccl_restrict variance_image,
+                                                         float *difference_image,
+                                                         int4 rect,
+                                                         int w,
+                                                         int channel_offset,
+                                                         float a,
+                                                         float k_2)
 {
 	for(int y = rect.y; y < rect.w; y++) {
 		for(int x = rect.x; x < rect.z; x++) {
 			float diff = 0.0f;
 			int numChannels = channel_offset? 3 : 1;
 			for(int c = 0; c < numChannels; c++) {
-				float cdiff = weightImage[c*channel_offset + y*w+x] - weightImage[c*channel_offset + (y+dy)*w+(x+dx)];
-				float pvar = varianceImage[c*channel_offset + y*w+x];
-				float qvar = varianceImage[c*channel_offset + (y+dy)*w+(x+dx)];
+				float cdiff = weight_image[c*channel_offset + y*w+x] - weight_image[c*channel_offset + (y+dy)*w+(x+dx)];
+				float pvar = variance_image[c*channel_offset + y*w+x];
+				float qvar = variance_image[c*channel_offset + (y+dy)*w+(x+dx)];
 				diff += (cdiff*cdiff - a*(pvar + min(pvar, qvar))) / (1e-8f + k_2*(pvar+qvar));
 			}
 			if(numChannels > 1) {
 				diff *= 1.0f/numChannels;
 			}
-			differenceImage[y*w+x] = diff;
+			difference_image[y*w+x] = diff;
 		}
 	}
 }
 
-ccl_device_inline void kernel_filter_nlm_blur(float ccl_restrict_ptr differenceImage, float *outImage, int4 rect, int w, int f)
+ccl_device_inline void kernel_filter_nlm_blur(const float *ccl_restrict difference_image,
+                                              float *out_image,
+                                              int4 rect,
+                                              int w,
+                                              int f)
 {
 #ifdef __KERNEL_SSE3__
 	int aligned_lowx = (rect.x & ~(3));
@@ -46,30 +58,34 @@ ccl_device_inline void kernel_filter_nlm_blur(float ccl_restrict_ptr differenceI
 		const int low = max(rect.y, y-f);
 		const int high = min(rect.w, y+f+1);
 		for(int x = rect.x; x < rect.z; x++) {
-			outImage[y*w+x] = 0.0f;
+			out_image[y*w+x] = 0.0f;
 		}
 		for(int y1 = low; y1 < high; y1++) {
 #ifdef __KERNEL_SSE3__
 			for(int x = aligned_lowx; x < aligned_highx; x+=4) {
-				_mm_store_ps(outImage + y*w+x, _mm_add_ps(_mm_load_ps(outImage + y*w+x), _mm_load_ps(differenceImage + y1*w+x)));
+				_mm_store_ps(out_image + y*w+x, _mm_add_ps(_mm_load_ps(out_image + y*w+x), _mm_load_ps(difference_image + y1*w+x)));
 			}
 #else
 			for(int x = rect.x; x < rect.z; x++) {
-				outImage[y*w+x] += differenceImage[y1*w+x];
+				out_image[y*w+x] += difference_image[y1*w+x];
 			}
 #endif
 		}
 		for(int x = rect.x; x < rect.z; x++) {
-			outImage[y*w+x] *= 1.0f/(high - low);
+			out_image[y*w+x] *= 1.0f/(high - low);
 		}
 	}
 }
 
-ccl_device_inline void kernel_filter_nlm_calc_weight(float ccl_restrict_ptr differenceImage, float *outImage, int4 rect, int w, int f)
+ccl_device_inline void kernel_filter_nlm_calc_weight(const float *ccl_restrict difference_image,
+                                                     float *out_image,
+                                                     int4 rect,
+                                                     int w,
+                                                     int f)
 {
 	for(int y = rect.y; y < rect.w; y++) {
 		for(int x = rect.x; x < rect.z; x++) {
-			outImage[y*w+x] = 0.0f;
+			out_image[y*w+x] = 0.0f;
 		}
 	}
 	for(int dx = -f; dx <= f; dx++) {
@@ -77,7 +93,7 @@ ccl_device_inline void kernel_filter_nlm_calc_weight(float ccl_restrict_ptr diff
 		int neg_dx = min(0, dx);
 		for(int y = rect.y; y < rect.w; y++) {
 			for(int x = rect.x-neg_dx; x < rect.z-pos_dx; x++) {
-				outImage[y*w+x] += differenceImage[y*w+dx+x];
+				out_image[y*w+x] += difference_image[y*w+dx+x];
 			}
 		}
 	}
@@ -85,12 +101,19 @@ ccl_device_inline void kernel_filter_nlm_calc_weight(float ccl_restrict_ptr diff
 		for(int x = rect.x; x < rect.z; x++) {
 			const int low = max(rect.x, x-f);
 			const int high = min(rect.z, x+f+1);
-			outImage[y*w+x] = expf(-max(outImage[y*w+x] * (1.0f/(high - low)), 0.0f));
+			out_image[y*w+x] = fast_expf(-max(out_image[y*w+x] * (1.0f/(high - low)), 0.0f));
 		}
 	}
 }
 
-ccl_device_inline void kernel_filter_nlm_update_output(int dx, int dy, float ccl_restrict_ptr differenceImage, float ccl_restrict_ptr image, float *outImage, float *accumImage, int4 rect, int w, int f)
+ccl_device_inline void kernel_filter_nlm_update_output(int dx, int dy,
+                                                       const float *ccl_restrict difference_image,
+                                                       const float *ccl_restrict image,
+                                                       float *out_image,
+                                                       float *accum_image,
+                                                       int4 rect,
+                                                       int w,
+                                                       int f)
 {
 	for(int y = rect.y; y < rect.w; y++) {
 		for(int x = rect.x; x < rect.z; x++) {
@@ -98,20 +121,18 @@ ccl_device_inline void kernel_filter_nlm_update_output(int dx, int dy, float ccl
 			const int high = min(rect.z, x+f+1);
 			float sum = 0.0f;
 			for(int x1 = low; x1 < high; x1++) {
-				sum += differenceImage[y*w+x1];
+				sum += difference_image[y*w+x1];
 			}
 			float weight = sum * (1.0f/(high - low));
-			accumImage[y*w+x] += weight;
-			outImage[y*w+x] += weight*image[(y+dy)*w+(x+dx)];
+			accum_image[y*w+x] += weight;
+			out_image[y*w+x] += weight*image[(y+dy)*w+(x+dx)];
 		}
 	}
 }
 
 ccl_device_inline void kernel_filter_nlm_construct_gramian(int dx, int dy,
-                                                           float ccl_restrict_ptr differenceImage,
-                                                           float ccl_restrict_ptr buffer,
-                                                           float *color_pass,
-                                                           float *variance_pass,
+                                                           const float *ccl_restrict difference_image,
+                                                           const float *ccl_restrict buffer,
                                                            float *transform,
                                                            int *rank,
                                                            float *XtWX,
@@ -130,7 +151,7 @@ ccl_device_inline void kernel_filter_nlm_construct_gramian(int dx, int dy,
 			const int high = min(rect.z, x+f+1);
 			float sum = 0.0f;
 			for(int x1 = low; x1 < high; x1++) {
-				sum += differenceImage[y*w+x1];
+				sum += difference_image[y*w+x1];
 			}
 			float weight = sum * (1.0f/(high - low));
 
@@ -144,18 +165,20 @@ ccl_device_inline void kernel_filter_nlm_construct_gramian(int dx, int dy,
 			                                dx, dy, w, h,
 			                                pass_stride,
 			                                buffer,
-			                                color_pass, variance_pass,
 			                                l_transform, l_rank,
 			                                weight, l_XtWX, l_XtWY, 0);
 		}
 	}
 }
 
-ccl_device_inline void kernel_filter_nlm_normalize(float *outImage, float ccl_restrict_ptr accumImage, int4 rect, int w)
+ccl_device_inline void kernel_filter_nlm_normalize(float *out_image,
+                                                   const float *ccl_restrict accum_image,
+                                                   int4 rect,
+                                                   int w)
 {
 	for(int y = rect.y; y < rect.w; y++) {
 		for(int x = rect.x; x < rect.z; x++) {
-			outImage[y*w+x] /= accumImage[y*w+x];
+			out_image[y*w+x] /= accum_image[y*w+x];
 		}
 	}
 }

intern/cycles/kernel/filter/filter_nlm_gpu.h

@@ -18,9 +18,9 @@ CCL_NAMESPACE_BEGIN
 
 ccl_device_inline void kernel_filter_nlm_calc_difference(int x, int y,
                                                          int dx, int dy,
-                                                         ccl_global float ccl_restrict_ptr weightImage,
-                                                         ccl_global float ccl_restrict_ptr varianceImage,
-                                                         ccl_global float *differenceImage,
+                                                         const ccl_global float *ccl_restrict weight_image,
+                                                         const ccl_global float *ccl_restrict variance_image,
+                                                         ccl_global float *difference_image,
                                                          int4 rect, int w,
                                                          int channel_offset,
                                                          float a, float k_2)
@@ -28,78 +28,76 @@ ccl_device_inline void kernel_filter_nlm_calc_difference(int x, int y,
 	float diff = 0.0f;
 	int numChannels = channel_offset? 3 : 1;
 	for(int c = 0; c < numChannels; c++) {
-		float cdiff = weightImage[c*channel_offset + y*w+x] - weightImage[c*channel_offset + (y+dy)*w+(x+dx)];
-		float pvar = varianceImage[c*channel_offset + y*w+x];
-		float qvar = varianceImage[c*channel_offset + (y+dy)*w+(x+dx)];
+		float cdiff = weight_image[c*channel_offset + y*w+x] - weight_image[c*channel_offset + (y+dy)*w+(x+dx)];
+		float pvar = variance_image[c*channel_offset + y*w+x];
+		float qvar = variance_image[c*channel_offset + (y+dy)*w+(x+dx)];
 		diff += (cdiff*cdiff - a*(pvar + min(pvar, qvar))) / (1e-8f + k_2*(pvar+qvar));
 	}
 	if(numChannels > 1) {
 		diff *= 1.0f/numChannels;
 	}
-	differenceImage[y*w+x] = diff;
+	difference_image[y*w+x] = diff;
 }
 
 ccl_device_inline void kernel_filter_nlm_blur(int x, int y,
-                                              ccl_global float ccl_restrict_ptr differenceImage,
-                                              ccl_global float *outImage,
+                                              const ccl_global float *ccl_restrict difference_image,
+                                              ccl_global float *out_image,
                                               int4 rect, int w, int f)
 {
 	float sum = 0.0f;
 	const int low = max(rect.y, y-f);
 	const int high = min(rect.w, y+f+1);
 	for(int y1 = low; y1 < high; y1++) {
-		sum += differenceImage[y1*w+x];
+		sum += difference_image[y1*w+x];
 	}
 	sum *= 1.0f/(high-low);
-	outImage[y*w+x] = sum;
+	out_image[y*w+x] = sum;
 }
 
 ccl_device_inline void kernel_filter_nlm_calc_weight(int x, int y,
-                                                     ccl_global float ccl_restrict_ptr differenceImage,
-                                                     ccl_global float *outImage,
+                                                     const ccl_global float *ccl_restrict difference_image,
+                                                     ccl_global float *out_image,
                                                      int4 rect, int w, int f)
 {
 	float sum = 0.0f;
 	const int low = max(rect.x, x-f);
 	const int high = min(rect.z, x+f+1);
 	for(int x1 = low; x1 < high; x1++) {
-		sum += differenceImage[y*w+x1];
+		sum += difference_image[y*w+x1];
 	}
 	sum *= 1.0f/(high-low);
-	outImage[y*w+x] = expf(-max(sum, 0.0f));
+	out_image[y*w+x] = fast_expf(-max(sum, 0.0f));
 }
 
 ccl_device_inline void kernel_filter_nlm_update_output(int x, int y,
                                                        int dx, int dy,
-                                                       ccl_global float ccl_restrict_ptr differenceImage,
-                                                       ccl_global float ccl_restrict_ptr image,
-                                                       ccl_global float *outImage,
-                                                       ccl_global float *accumImage,
+                                                       const ccl_global float *ccl_restrict difference_image,
+                                                       const ccl_global float *ccl_restrict image,
+                                                       ccl_global float *out_image,
+                                                       ccl_global float *accum_image,
                                                        int4 rect, int w, int f)
 {
 	float sum = 0.0f;
 	const int low = max(rect.x, x-f);
 	const int high = min(rect.z, x+f+1);
 	for(int x1 = low; x1 < high; x1++) {
-		sum += differenceImage[y*w+x1];
+		sum += difference_image[y*w+x1];
 	}
 	sum *= 1.0f/(high-low);
-	if(outImage) {
-		accumImage[y*w+x] += sum;
-		outImage[y*w+x] += sum*image[(y+dy)*w+(x+dx)];
+	if(out_image) {
+		accum_image[y*w+x] += sum;
+		out_image[y*w+x] += sum*image[(y+dy)*w+(x+dx)];
 	}
 	else {
-		accumImage[y*w+x] = sum;
+		accum_image[y*w+x] = sum;
 	}
 }
 
 ccl_device_inline void kernel_filter_nlm_construct_gramian(int fx, int fy,
                                                            int dx, int dy,
-                                                           ccl_global float ccl_restrict_ptr differenceImage,
-                                                           ccl_global float ccl_restrict_ptr buffer,
-                                                           ccl_global float *color_pass,
-                                                           ccl_global float *variance_pass,
-                                                           ccl_global float ccl_restrict_ptr transform,
+                                                           const ccl_global float *ccl_restrict difference_image,
+                                                           const ccl_global float *ccl_restrict buffer,
+                                                           const ccl_global float *ccl_restrict transform,
                                                            ccl_global int *rank,
                                                            ccl_global float *XtWX,
                                                            ccl_global float3 *XtWY,
@@ -115,7 +113,7 @@ ccl_device_inline void kernel_filter_nlm_construct_gramian(int fx, int fy,
 	const int high = min(rect.z, x+f+1);
 	float sum = 0.0f;
 	for(int x1 = low; x1 < high; x1++) {
-		sum += differenceImage[y*w+x1];
+		sum += difference_image[y*w+x1];
 	}
 	float weight = sum * (1.0f/(high - low));
 
@@ -130,18 +128,17 @@ ccl_device_inline void kernel_filter_nlm_construct_gramian(int fx, int fy,
 	                                dx, dy, w, h,
 	                                pass_stride,
 	                                buffer,
-	                                color_pass, variance_pass,
 	                                transform, rank,
 	                                weight, XtWX, XtWY,
 	                                localIdx);
 }
 
 ccl_device_inline void kernel_filter_nlm_normalize(int x, int y,
-                                                   ccl_global float *outImage,
-                                                   ccl_global float ccl_restrict_ptr accumImage,
+                                                   ccl_global float *out_image,
+                                                   const ccl_global float *ccl_restrict accum_image,
                                                    int4 rect, int w)
 {
-	outImage[y*w+x] /= accumImage[y*w+x];
+	out_image[y*w+x] /= accum_image[y*w+x];
 }
 
 CCL_NAMESPACE_END

intern/cycles/kernel/filter/filter_prefilter.h

@@ -44,7 +44,7 @@ ccl_device void kernel_filter_divide_shadow(int sample,
 
 	int offset = tiles->offsets[tile];
 	int stride = tiles->strides[tile];
-	ccl_global float ccl_restrict_ptr center_buffer = (ccl_global float*) tiles->buffers[tile];
+	const ccl_global float *ccl_restrict center_buffer = (ccl_global float*) tiles->buffers[tile];
 	center_buffer += (y*stride + x + offset)*buffer_pass_stride;
 	center_buffer += buffer_denoising_offset + 14;
 
@@ -61,8 +61,8 @@ ccl_device void kernel_filter_divide_shadow(int sample,
 		varA = max(0.0f, varA - unfilteredA[idx]*unfilteredA[idx]*odd_sample);
 		varB = max(0.0f, varB - unfilteredB[idx]*unfilteredB[idx]*even_sample);
 	}
-	varA /= (odd_sample - 1);
-	varB /= (even_sample - 1);
+	varA /= max(odd_sample - 1, 1);
+	varB /= max(even_sample - 1, 1);
 
 	sampleVariance[idx]  = 0.5f*(varA + varB) / sample;
 	sampleVarianceV[idx] = 0.5f * (varA - varB) * (varA - varB) / (sample*sample);
@@ -77,7 +77,7 @@ ccl_device void kernel_filter_divide_shadow(int sample,
  * - mean, variance: Target denoise buffers.
  * - rect: The prefilter area (lower pixels inclusive, upper pixels exclusive).
  */
-ccl_device void kernel_filter_get_feature(int sample, 
+ccl_device void kernel_filter_get_feature(int sample,
                                           ccl_global TilesInfo *tiles,
                                           int m_offset, int v_offset,
                                           int x, int y,
@@ -96,14 +96,87 @@ ccl_device void kernel_filter_get_feature(int sample,
 	int idx = (y-rect.y)*buffer_w + (x - rect.x);
 
 	mean[idx] = center_buffer[m_offset] / sample;
-	if(use_split_variance) {
-		variance[idx] = max(0.0f, (center_buffer[v_offset] - mean[idx]*mean[idx]*sample) / (sample * (sample-1)));
+	if (sample > 1) {
+		if(use_split_variance) {
+			variance[idx] = max(0.0f, (center_buffer[v_offset] - mean[idx]*mean[idx]*sample) / (sample * (sample-1)));
+		}
+		else {
+			variance[idx] = center_buffer[v_offset] / (sample * (sample-1));
+		}
 	}
 	else {
-		variance[idx] = center_buffer[v_offset] / (sample * (sample-1));
+		/* Can't compute variance with single sample, just set it very high. */
+		variance[idx] = 1e10f;
 	}
 }
 
+ccl_device void kernel_filter_detect_outliers(int x, int y,
+                                              ccl_global float *image,
+                                              ccl_global float *variance,
+                                              ccl_global float *depth,
+                                              ccl_global float *out,
+                                              int4 rect,
+                                              int pass_stride)
+{
+	int buffer_w = align_up(rect.z - rect.x, 4);
+
+	int idx = (y-rect.y)*buffer_w + (x-rect.x);
+	float3 color = make_float3(image[idx], image[idx+pass_stride], image[idx+2*pass_stride]);
+
+	float fac = 1.0f;
+	if(color.x < 0.0f || color.y < 0.0f || color.z < 0.0f) {
+		depth[idx] = -depth[idx];
+		fac = 0.0f;
+	}
+	else {
+		float L = average(color);
+		int n = 0;
+		float values[25];
+		for(int y1 = max(y-2, rect.y); y1 < min(y+3, rect.w); y1++) {
+			for(int x1 = max(x-2, rect.x); x1 < min(x+3, rect.z); x1++) {
+				int idx = (y1-rect.y)*buffer_w + (x1-rect.x);
+				float L = average(make_float3(image[idx], image[idx+pass_stride], image[idx+2*pass_stride]));
+
+				/* Find the position of L. */
+				int i;
+				for(i = 0; i < n; i++) {
+					if(values[i] > L) break;
+				}
+				/* Make space for L by shifting all following values to the right. */
+				for(int j = n; j > i; j--) {
+					values[j] = values[j-1];
+				}
+				/* Insert L. */
+				values[i] = L;
+				n++;
+			}
+		}
+
+		float ref = 2.0f*values[(int)(n*0.75f)];
+		if(L > ref) {
+			/* The pixel appears to be an outlier.
+			 * However, it may just be a legitimate highlight. Therefore, it is checked how likely it is that the pixel
+			 * should actually be at the reference value:
+			 * If the reference is within the 3-sigma interval, the pixel is assumed to be a statistical outlier.
+			 * Otherwise, it is very unlikely that the pixel should be darker, which indicates a legitimate highlight.
+			 */
+			float stddev = sqrtf(average(make_float3(variance[idx], variance[idx+pass_stride], variance[idx+2*pass_stride])));
+			if(L - 3*stddev < ref) {
+				/* The pixel is an outlier, so negate the depth value to mark it as one.
+				 * Also, scale its brightness down to the outlier threshold to avoid trouble with the NLM weights. */
+				depth[idx] = -depth[idx];
+				fac = ref/L;
+				variance[idx              ] *= fac*fac;
+				variance[idx + pass_stride] *= fac*fac;
+				variance[idx+2*pass_stride] *= fac*fac;
+			}
+		}
+	}
+	out[idx              ] = fac*image[idx];
+	out[idx + pass_stride] = fac*image[idx + pass_stride];
+	out[idx+2*pass_stride] = fac*image[idx+2*pass_stride];
+}
+
 /* Combine A/B buffers.
  * Calculates the combined mean and the buffer variance. */
 ccl_device void kernel_filter_combine_halves(int x, int y,

intern/cycles/kernel/filter/filter_reconstruction.h

@@ -21,40 +21,39 @@ ccl_device_inline void kernel_filter_construct_gramian(int x, int y,
                                                        int dx, int dy,
                                                        int w, int h,
                                                        int pass_stride,
-                                                       ccl_global float ccl_restrict_ptr buffer,
-                                                       ccl_global float *color_pass,
-                                                       ccl_global float *variance_pass,
-                                                       ccl_global float ccl_restrict_ptr transform,
+                                                       const ccl_global float *ccl_restrict buffer,
+                                                       const ccl_global float *ccl_restrict transform,
                                                        ccl_global int *rank,
                                                        float weight,
                                                        ccl_global float *XtWX,
                                                        ccl_global float3 *XtWY,
                                                        int localIdx)
 {
+	if(weight < 1e-3f) {
+		return;
+	}
+
 	int p_offset =  y    *w +  x;
 	int q_offset = (y+dy)*w + (x+dx);
 
-#ifdef __KERNEL_CPU__
-	const int stride = 1;
-	(void)storage_stride;
-	(void)localIdx;
-	float design_row[DENOISE_FEATURES+1];
-#elif defined(__KERNEL_CUDA__)
+#ifdef __KERNEL_GPU__
 	const int stride = storage_stride;
+#else
+	const int stride = 1;
+	(void) storage_stride;
+#endif
+
+#ifdef __KERNEL_CUDA__
 	ccl_local float shared_design_row[(DENOISE_FEATURES+1)*CCL_MAX_LOCAL_SIZE];
 	ccl_local_param float *design_row = shared_design_row + localIdx*(DENOISE_FEATURES+1);
 #else
-	const int stride = storage_stride;
 	float design_row[DENOISE_FEATURES+1];
 #endif
 
-	float3 p_color = filter_get_pixel_color(color_pass + p_offset, pass_stride);
-	float3 q_color = filter_get_pixel_color(color_pass + q_offset, pass_stride);
+	float3 q_color = filter_get_color(buffer + q_offset, pass_stride);
 
-	float p_std_dev = sqrtf(filter_get_pixel_variance(variance_pass + p_offset, pass_stride));
-	float q_std_dev = sqrtf(filter_get_pixel_variance(variance_pass + q_offset, pass_stride));
-
-	if(average(fabs(p_color - q_color)) > 3.0f*(p_std_dev + q_std_dev + 1e-3f)) {
+	/* If the pixel was flagged as an outlier during prefiltering, skip it. */
+	if(ccl_get_feature(buffer + q_offset, 0) < 0.0f) {
 		return;
 	}
 
@@ -75,16 +74,33 @@ ccl_device_inline void kernel_filter_finalize(int x, int y, int w, int h,
                                               int4 buffer_params,
                                               int sample)
 {
-#ifdef __KERNEL_CPU__
-	const int stride = 1;
-	(void)storage_stride;
-#else
+#ifdef __KERNEL_GPU__
 	const int stride = storage_stride;
+#else
+	const int stride = 1;
+	(void) storage_stride;
 #endif
 
+	if(XtWX[0] < 1e-3f) {
+		/* There is not enough information to determine a denoised result.
+		 * As a fallback, keep the original value of the pixel. */
+		 return;
+	}
+
+	/* The weighted average of pixel colors (essentially, the NLM-filtered image).
+	 * In case the solution of the linear model fails due to numerical issues,
+	 * fall back to this value. */
+	float3 mean_color = XtWY[0]/XtWX[0];
+
 	math_trimatrix_vec3_solve(XtWX, XtWY, (*rank)+1, stride);
 
 	float3 final_color = XtWY[0];
+	if(!isfinite3_safe(final_color)) {
+		final_color = mean_color;
+	}
+
+	/* Clamp pixel value to positive values. */
+	final_color = max(final_color, make_float3(0.0f, 0.0f, 0.0f));
 
 	ccl_global float *combined_buffer = buffer + (y*buffer_params.y + x + buffer_params.x)*buffer_params.z;
 	final_color *= sample;
@@ -98,6 +114,4 @@ ccl_device_inline void kernel_filter_finalize(int x, int y, int w, int h,
 	combined_buffer[2] = final_color.z;
 }
 
-#undef STORAGE_TYPE
-
 CCL_NAMESPACE_END

intern/cycles/kernel/filter/filter_transform.h

@@ -16,7 +16,7 @@
 
 CCL_NAMESPACE_BEGIN
 
-ccl_device void kernel_filter_construct_transform(float ccl_restrict_ptr buffer,
+ccl_device void kernel_filter_construct_transform(const float *ccl_restrict buffer,
                                                   int x, int y, int4 rect,
                                                   int pass_stride,
                                                   float *transform, int *rank,
@@ -29,20 +29,15 @@ ccl_device void kernel_filter_construct_transform(float ccl_restrict_ptr buffer,
 	/* Temporary storage, used in different steps of the algorithm. */
 	float tempmatrix[DENOISE_FEATURES*DENOISE_FEATURES];
 	float tempvector[2*DENOISE_FEATURES];
-	float ccl_restrict_ptr pixel_buffer;
+	const float *ccl_restrict pixel_buffer;
 	int2 pixel;
 
-
-
-
 	/* === Calculate denoising window. === */
 	int2 low  = make_int2(max(rect.x, x - radius),
 	                      max(rect.y, y - radius));
 	int2 high = make_int2(min(rect.z, x + radius + 1),
 	                      min(rect.w, y + radius + 1));
-
-
-
+	int num_pixels = (high.y - low.y) * (high.x - low.x);
 
 	/* === Shift feature passes to have mean 0. === */
 	float feature_means[DENOISE_FEATURES];
@@ -52,8 +47,7 @@ ccl_device void kernel_filter_construct_transform(float ccl_restrict_ptr buffer,
 		math_vector_add(feature_means, features, DENOISE_FEATURES);
 	} END_FOR_PIXEL_WINDOW
 
-	float pixel_scale = 1.0f / ((high.y - low.y) * (high.x - low.x));
-	math_vector_scale(feature_means, pixel_scale, DENOISE_FEATURES);
+	math_vector_scale(feature_means, 1.0f / num_pixels, DENOISE_FEATURES);
 
 	/* === Scale the shifted feature passes to a range of [-1; 1], will be baked into the transform later. === */
 	float *feature_scale = tempvector;
@@ -66,7 +60,6 @@ ccl_device void kernel_filter_construct_transform(float ccl_restrict_ptr buffer,
 
 	filter_calculate_scale(feature_scale);
 
-
 	/* === Generate the feature transformation. ===
 	 * This transformation maps the DENOISE_FEATURES-dimentional feature space to a reduced feature (r-feature) space
 	 * which generally has fewer dimensions. This mainly helps to prevent overfitting. */
@@ -80,6 +73,8 @@ ccl_device void kernel_filter_construct_transform(float ccl_restrict_ptr buffer,
 
 	math_matrix_jacobi_eigendecomposition(feature_matrix, transform, DENOISE_FEATURES, 1);
 	*rank = 0;
+	/* Prevent overfitting when a small window is used. */
+	int max_rank = min(DENOISE_FEATURES, num_pixels/3);
 	if(pca_threshold < 0.0f) {
 		float threshold_energy = 0.0f;
 		for(int i = 0; i < DENOISE_FEATURES; i++) {
@@ -88,24 +83,25 @@ ccl_device void kernel_filter_construct_transform(float ccl_restrict_ptr buffer,
 		threshold_energy *= 1.0f - (-pca_threshold);
 
 		float reduced_energy = 0.0f;
-		for(int i = 0; i < DENOISE_FEATURES; i++, (*rank)++) {
+		for(int i = 0; i < max_rank; i++, (*rank)++) {
 			if(i >= 2 && reduced_energy >= threshold_energy)
 				break;
 			float s = feature_matrix[i*DENOISE_FEATURES+i];
 			reduced_energy += s;
-			/* Bake the feature scaling into the transformation matrix. */
-			math_vector_mul(transform + (*rank)*DENOISE_FEATURES, feature_scale, DENOISE_FEATURES);
 		}
 	}
 	else {
-		for(int i = 0; i < DENOISE_FEATURES; i++, (*rank)++) {
+		for(int i = 0; i < max_rank; i++, (*rank)++) {
 			float s = feature_matrix[i*DENOISE_FEATURES+i];
 			if(i >= 2 && sqrtf(s) < pca_threshold)
 				break;
-			/* Bake the feature scaling into the transformation matrix. */
-			math_vector_mul(transform + (*rank)*DENOISE_FEATURES, feature_scale, DENOISE_FEATURES);
 		}
 	}
+
+	/* Bake the feature scaling into the transformation matrix. */
+	for(int i = 0; i < (*rank); i++) {
+		math_vector_mul(transform + i*DENOISE_FEATURES, feature_scale, DENOISE_FEATURES);
+	}
 	math_matrix_transpose(transform, DENOISE_FEATURES, 1);
 }
 

intern/cycles/kernel/filter/filter_transform_gpu.h

@@ -16,7 +16,7 @@
 
 CCL_NAMESPACE_BEGIN
 
-ccl_device void kernel_filter_construct_transform(ccl_global float ccl_restrict_ptr buffer,
+ccl_device void kernel_filter_construct_transform(const ccl_global float *ccl_restrict buffer,
                                                   int x, int y, int4 rect,
                                                   int pass_stride,
                                                   ccl_global float *transform,
@@ -38,7 +38,8 @@ ccl_device void kernel_filter_construct_transform(ccl_global float ccl_restrict_
 	                      max(rect.y, y - radius));
 	int2 high = make_int2(min(rect.z, x + radius + 1),
 	                      min(rect.w, y + radius + 1));
-	ccl_global float ccl_restrict_ptr pixel_buffer;
+	int num_pixels = (high.y - low.y) * (high.x - low.x);
+	const ccl_global float *ccl_restrict pixel_buffer;
 	int2 pixel;
 
 
@@ -52,8 +53,7 @@ ccl_device void kernel_filter_construct_transform(ccl_global float ccl_restrict_
 		math_vector_add(feature_means, features, DENOISE_FEATURES);
 	} END_FOR_PIXEL_WINDOW
 
-	float pixel_scale = 1.0f / ((high.y - low.y) * (high.x - low.x));
-	math_vector_scale(feature_means, pixel_scale, DENOISE_FEATURES);
+	math_vector_scale(feature_means, 1.0f / num_pixels, DENOISE_FEATURES);
 
 	/* === Scale the shifted feature passes to a range of [-1; 1], will be baked into the transform later. === */
 	float feature_scale[DENOISE_FEATURES];
@@ -81,6 +81,8 @@ ccl_device void kernel_filter_construct_transform(ccl_global float ccl_restrict_
 
 	math_matrix_jacobi_eigendecomposition(feature_matrix, transform, DENOISE_FEATURES, transform_stride);
 	*rank = 0;
+	/* Prevent overfitting when a small window is used. */
+	int max_rank = min(DENOISE_FEATURES, num_pixels/3);
 	if(pca_threshold < 0.0f) {
 		float threshold_energy = 0.0f;
 		for(int i = 0; i < DENOISE_FEATURES; i++) {
@@ -89,25 +91,29 @@ ccl_device void kernel_filter_construct_transform(ccl_global float ccl_restrict_
 		threshold_energy *= 1.0f - (-pca_threshold);
 
 		float reduced_energy = 0.0f;
-		for(int i = 0; i < DENOISE_FEATURES; i++, (*rank)++) {
+		for(int i = 0; i < max_rank; i++, (*rank)++) {
 			if(i >= 2 && reduced_energy >= threshold_energy)
 				break;
 			float s = feature_matrix[i*DENOISE_FEATURES+i];
 			reduced_energy += s;
-			/* Bake the feature scaling into the transformation matrix. */
-			math_vector_mul_strided(transform + i*DENOISE_FEATURES*transform_stride, feature_scale, transform_stride, DENOISE_FEATURES);
 		}
 	}
 	else {
-		for(int i = 0; i < DENOISE_FEATURES; i++, (*rank)++) {
+		for(int i = 0; i < max_rank; i++, (*rank)++) {
 			float s = feature_matrix[i*DENOISE_FEATURES+i];
 			if(i >= 2 && sqrtf(s) < pca_threshold)
 				break;
-			/* Bake the feature scaling into the transformation matrix. */
-			math_vector_mul_strided(transform + i*DENOISE_FEATURES*transform_stride, feature_scale, transform_stride, DENOISE_FEATURES);
 		}
 	}
+
 	math_matrix_transpose(transform, DENOISE_FEATURES, transform_stride);
+
+	/* Bake the feature scaling into the transformation matrix. */
+	for(int i = 0; i < DENOISE_FEATURES; i++) {
+		for(int j = 0; j < (*rank); j++) {
+			transform[(i*DENOISE_FEATURES + j)*transform_stride] *= feature_scale[i];
+		}
+	}
 }
 
 CCL_NAMESPACE_END

intern/cycles/kernel/filter/filter_transform_sse.h

@@ -16,7 +16,7 @@
 
 CCL_NAMESPACE_BEGIN
 
-ccl_device void kernel_filter_construct_transform(float ccl_restrict_ptr buffer,
+ccl_device void kernel_filter_construct_transform(const float *ccl_restrict buffer,
                                                   int x, int y, int4 rect,
                                                   int pass_stride,
                                                   float *transform, int *rank,
@@ -25,13 +25,14 @@ ccl_device void kernel_filter_construct_transform(float ccl_restrict_ptr buffer,
 	int buffer_w = align_up(rect.z - rect.x, 4);
 
 	__m128 features[DENOISE_FEATURES];
-	float ccl_restrict_ptr pixel_buffer;
+	const float *ccl_restrict pixel_buffer;
 	int2 pixel;
 
 	int2 low  = make_int2(max(rect.x, x - radius),
 	                      max(rect.y, y - radius));
 	int2 high = make_int2(min(rect.z, x + radius + 1),
 	                      min(rect.w, y + radius + 1));
+	int num_pixels = (high.y - low.y) * (high.x - low.x);
 
 	__m128 feature_means[DENOISE_FEATURES];
 	math_vector_zero_sse(feature_means, DENOISE_FEATURES);
@@ -40,7 +41,7 @@ ccl_device void kernel_filter_construct_transform(float ccl_restrict_ptr buffer,
 		math_vector_add_sse(feature_means, DENOISE_FEATURES, features);
 	} END_FOR_PIXEL_WINDOW_SSE
 
-	__m128 pixel_scale = _mm_set1_ps(1.0f / ((high.y - low.y) * (high.x - low.x)));
+	__m128 pixel_scale = _mm_set1_ps(1.0f / num_pixels);
 	for(int i = 0; i < DENOISE_FEATURES; i++) {
 		feature_means[i] = _mm_mul_ps(_mm_hsum_ps(feature_means[i]), pixel_scale);
 	}
@@ -49,8 +50,7 @@ ccl_device void kernel_filter_construct_transform(float ccl_restrict_ptr buffer,
 	math_vector_zero_sse(feature_scale, DENOISE_FEATURES);
 	FOR_PIXEL_WINDOW_SSE {
 		filter_get_feature_scales_sse(x4, y4, active_pixels, pixel_buffer, features, feature_means, pass_stride);
-		for(int i = 0; i < DENOISE_FEATURES; i++)
-			feature_scale[i] = _mm_max_ps(feature_scale[i], features[i]);
+		math_vector_max_sse(feature_scale, features, DENOISE_FEATURES);
 	} END_FOR_PIXEL_WINDOW_SSE
 
 	filter_calculate_scale_sse(feature_scale);
@@ -69,6 +69,8 @@ ccl_device void kernel_filter_construct_transform(float ccl_restrict_ptr buffer,
 	math_matrix_jacobi_eigendecomposition(feature_matrix, transform, DENOISE_FEATURES, 1);
 
 	*rank = 0;
+	/* Prevent overfitting when a small window is used. */
+	int max_rank = min(DENOISE_FEATURES, num_pixels/3);
 	if(pca_threshold < 0.0f) {
 		float threshold_energy = 0.0f;
 		for(int i = 0; i < DENOISE_FEATURES; i++) {
@@ -77,29 +79,27 @@ ccl_device void kernel_filter_construct_transform(float ccl_restrict_ptr buffer,
 		threshold_energy *= 1.0f - (-pca_threshold);
 
 		float reduced_energy = 0.0f;
-		for(int i = 0; i < DENOISE_FEATURES; i++, (*rank)++) {
+		for(int i = 0; i < max_rank; i++, (*rank)++) {
 			if(i >= 2 && reduced_energy >= threshold_energy)
 				break;
 			float s = feature_matrix[i*DENOISE_FEATURES+i];
 			reduced_energy += s;
-			/* Bake the feature scaling into the transformation matrix. */
-			for(int j = 0; j < DENOISE_FEATURES; j++) {
-				transform[(*rank)*DENOISE_FEATURES + j] *= _mm_cvtss_f32(feature_scale[j]);
-			}
 		}
 	}
 	else {
-		for(int i = 0; i < DENOISE_FEATURES; i++, (*rank)++) {
+		for(int i = 0; i < max_rank; i++, (*rank)++) {
 			float s = feature_matrix[i*DENOISE_FEATURES+i];
 			if(i >= 2 && sqrtf(s) < pca_threshold)
 				break;
-			/* Bake the feature scaling into the transformation matrix. */
-			for(int j = 0; j < DENOISE_FEATURES; j++) {
-				transform[(*rank)*DENOISE_FEATURES + j] *= _mm_cvtss_f32(feature_scale[j]);
-			}
 		}
 	}
+
 	math_matrix_transpose(transform, DENOISE_FEATURES, 1);
+
+	/* Bake the feature scaling into the transformation matrix. */
+	for(int i = 0; i < DENOISE_FEATURES; i++) {
+		math_vector_scale(transform + i*DENOISE_FEATURES, _mm_cvtss_f32(feature_scale[i]), *rank);
+	}
 }
 
 CCL_NAMESPACE_END

intern/cycles/kernel/kernel_accumulate.h

@@ -220,7 +220,9 @@ ccl_device_inline void path_radiance_init(PathRadiance *L, int use_light_pass)
 #ifdef __SHADOW_TRICKS__
 	L->path_total = make_float3(0.0f, 0.0f, 0.0f);
 	L->path_total_shaded = make_float3(0.0f, 0.0f, 0.0f);
-	L->shadow_color = make_float3(0.0f, 0.0f, 0.0f);
+	L->shadow_background_color = make_float3(0.0f, 0.0f, 0.0f);
+	L->shadow_radiance_sum = make_float3(0.0f, 0.0f, 0.0f);
+	L->shadow_throughput = 0.0f;
 #endif
 
 #ifdef __DENOISING_FEATURES__
@@ -621,25 +623,43 @@ ccl_device_inline void path_radiance_accum_sample(PathRadiance *L, PathRadiance
 {
 	float fac = 1.0f/num_samples;
 
+#ifdef __SPLIT_KERNEL__
+#  define safe_float3_add(f, v) \
+	do { \
+		ccl_global float *p = (ccl_global float*)(&(f)); \
+		atomic_add_and_fetch_float(p+0, (v).x); \
+		atomic_add_and_fetch_float(p+1, (v).y); \
+		atomic_add_and_fetch_float(p+2, (v).z); \
+	} while(0)
+#else
+#  define safe_float3_add(f, v) (f) += (v)
+#endif  /* __SPLIT_KERNEL__ */
+
 #ifdef __PASSES__
-	L->direct_diffuse += L_sample->direct_diffuse*fac;
-	L->direct_glossy += L_sample->direct_glossy*fac;
-	L->direct_transmission += L_sample->direct_transmission*fac;
-	L->direct_subsurface += L_sample->direct_subsurface*fac;
-	L->direct_scatter += L_sample->direct_scatter*fac;
+	safe_float3_add(L->direct_diffuse, L_sample->direct_diffuse*fac);
+	safe_float3_add(L->direct_glossy, L_sample->direct_glossy*fac);
+	safe_float3_add(L->direct_transmission, L_sample->direct_transmission*fac);
+	safe_float3_add(L->direct_subsurface, L_sample->direct_subsurface*fac);
+	safe_float3_add(L->direct_scatter, L_sample->direct_scatter*fac);
 
-	L->indirect_diffuse += L_sample->indirect_diffuse*fac;
-	L->indirect_glossy += L_sample->indirect_glossy*fac;
-	L->indirect_transmission += L_sample->indirect_transmission*fac;
-	L->indirect_subsurface += L_sample->indirect_subsurface*fac;
-	L->indirect_scatter += L_sample->indirect_scatter*fac;
+	safe_float3_add(L->indirect_diffuse, L_sample->indirect_diffuse*fac);
+	safe_float3_add(L->indirect_glossy, L_sample->indirect_glossy*fac);
+	safe_float3_add(L->indirect_transmission, L_sample->indirect_transmission*fac);
+	safe_float3_add(L->indirect_subsurface, L_sample->indirect_subsurface*fac);
+	safe_float3_add(L->indirect_scatter, L_sample->indirect_scatter*fac);
 
-	L->background += L_sample->background*fac;
-	L->ao += L_sample->ao*fac;
-	L->shadow += L_sample->shadow*fac;
+	safe_float3_add(L->background, L_sample->background*fac);
+	safe_float3_add(L->ao, L_sample->ao*fac);
+	safe_float3_add(L->shadow, L_sample->shadow*fac);
+#  ifdef __SPLIT_KERNEL__
+	atomic_add_and_fetch_float(&L->mist, L_sample->mist*fac);
+#  else
 	L->mist += L_sample->mist*fac;
-#endif
-	L->emission += L_sample->emission * fac;
+#  endif  /* __SPLIT_KERNEL__ */
+#endif  /* __PASSES__ */
+	safe_float3_add(L->emission, L_sample->emission*fac);
+
+#undef safe_float3_add
 }
 
 #ifdef __SHADOW_TRICKS__
@@ -662,11 +682,12 @@ ccl_device_inline float3 path_radiance_sum_shadowcatcher(KernelGlobals *kg,
 	const float shadow = path_radiance_sum_shadow(L);
 	float3 L_sum;
 	if(kernel_data.background.transparent) {
-		*alpha = 1.0f-shadow;
-		L_sum = make_float3(0.0f, 0.0f, 0.0f);
+		*alpha = 1.0f - L->shadow_throughput * shadow;
+		L_sum = L->shadow_radiance_sum;
 	}
 	else {
-		L_sum = L->shadow_color * shadow;
+		L_sum = L->shadow_background_color * L->shadow_throughput * shadow +
+		        L->shadow_radiance_sum;
 	}
 	return L_sum;
 }

intern/cycles/kernel/kernel_compat_cpu.h

@@ -42,8 +42,6 @@
 #include "util/util_types.h"
 #include "util/util_texture.h"
 
-#define ccl_restrict_ptr const * __restrict
-
 #define ccl_addr_space
 
 #define ccl_local_id(d) 0
@@ -197,7 +195,7 @@ template<typename T> struct texture_image  {
 					if(x < 0.0f || y < 0.0f || x > 1.0f || y > 1.0f) {
 						return make_float4(0.0f, 0.0f, 0.0f, 0.0f);
 					}
-					/* Fall through. */
+					ATTR_FALLTHROUGH;
 				case EXTENSION_EXTEND:
 					ix = wrap_clamp(ix, width);
 					iy = wrap_clamp(iy, height);
@@ -224,7 +222,7 @@ template<typename T> struct texture_image  {
 					if(x < 0.0f || y < 0.0f || x > 1.0f || y > 1.0f) {
 						return make_float4(0.0f, 0.0f, 0.0f, 0.0f);
 					}
-					/* Fall through. */
+					ATTR_FALLTHROUGH;
 				case EXTENSION_EXTEND:
 					nix = wrap_clamp(ix+1, width);
 					niy = wrap_clamp(iy+1, height);
@@ -267,7 +265,7 @@ template<typename T> struct texture_image  {
 					if(x < 0.0f || y < 0.0f || x > 1.0f || y > 1.0f) {
 						return make_float4(0.0f, 0.0f, 0.0f, 0.0f);
 					}
-					/* Fall through. */
+					ATTR_FALLTHROUGH;
 				case EXTENSION_EXTEND:
 					pix = wrap_clamp(ix-1, width);
 					piy = wrap_clamp(iy-1, height);
@@ -337,7 +335,7 @@ template<typename T> struct texture_image  {
 				{
 					return make_float4(0.0f, 0.0f, 0.0f, 0.0f);
 				}
-				/* Fall through. */
+				ATTR_FALLTHROUGH;
 			case EXTENSION_EXTEND:
 				ix = wrap_clamp(ix, width);
 				iy = wrap_clamp(iy, height);
@@ -376,7 +374,7 @@ template<typename T> struct texture_image  {
 				{
 					return make_float4(0.0f, 0.0f, 0.0f, 0.0f);
 				}
-				/* Fall through. */
+				ATTR_FALLTHROUGH;
 			case EXTENSION_EXTEND:
 				nix = wrap_clamp(ix+1, width);
 				niy = wrap_clamp(iy+1, height);
@@ -451,7 +449,7 @@ template<typename T> struct texture_image  {
 				{
 					return make_float4(0.0f, 0.0f, 0.0f, 0.0f);
 				}
-				/* Fall through. */
+				ATTR_FALLTHROUGH;
 			case EXTENSION_EXTEND:
 				pix = wrap_clamp(ix-1, width);
 				piy = wrap_clamp(iy-1, height);

intern/cycles/kernel/kernel_compat_cuda.h

@@ -55,7 +55,8 @@
 #define ccl_restrict __restrict__
 #define ccl_align(n) __align__(n)
 
-#define ccl_restrict_ptr const * __restrict__
+#define ATTR_FALLTHROUGH
+
 #define CCL_MAX_LOCAL_SIZE (CUDA_THREADS_BLOCK_WIDTH*CUDA_THREADS_BLOCK_WIDTH)
 
 

intern/cycles/kernel/kernel_compat_opencl.h

@@ -50,7 +50,7 @@
 #  define ccl_addr_space
 #endif
 
-#define ccl_restrict_ptr const * __restrict__
+#define ATTR_FALLTHROUGH
 
 #define ccl_local_id(d) get_local_id(d)
 #define ccl_global_id(d) get_global_id(d)

intern/cycles/kernel/kernel_jitter.h

@@ -175,15 +175,26 @@ ccl_device float cmj_sample_1D(int s, int N, int p)
 	return (x + jx)*invN;
 }
 
+/* TODO(sergey): Do some extra tests and consider moving to util_math.h. */
+ccl_device_inline int cmj_isqrt(int value)
+{
+#if defined(__KERNEL_CUDA__)
+	return float_to_int(__fsqrt_ru(value));
+#elif defined(__KERNEL_GPU__)
+	return float_to_int(sqrtf(value));
+#else
+	/* This is a work around for fast-math on CPU which might replace sqrtf()
+	 * with am approximated version.
+	 */
+	return float_to_int(sqrtf(value) + 1e-6f);
+#endif
+}
+
 ccl_device void cmj_sample_2D(int s, int N, int p, float *fx, float *fy)
 {
 	kernel_assert(s < N);
 
-#if defined(__KERNEL_CUDA__)
-	int m = float_to_int(__fsqrt_ru(N));
-#else
-	int m = float_to_int(sqrtf(N));
-#endif
+	int m = cmj_isqrt(N);
 	int n = (N - 1)/m + 1;
 	float invN = 1.0f/N;
 	float invm = 1.0f/m;

intern/cycles/kernel/kernel_passes.h

@@ -142,7 +142,7 @@ ccl_device_inline void kernel_write_denoising_shadow(KernelGlobals *kg, ccl_glob
 
 ccl_device_inline void kernel_update_denoising_features(KernelGlobals *kg,
                                                         ShaderData *sd,
-                                                        ccl_global PathState *state,
+                                                        ccl_addr_space PathState *state,
                                                         PathRadiance *L)
 {
 #ifdef __DENOISING_FEATURES__
@@ -152,6 +152,11 @@ ccl_device_inline void kernel_update_denoising_features(KernelGlobals *kg,
 
 	L->denoising_depth += ensure_finite(state->denoising_feature_weight * sd->ray_length);
 
+	/* Skip implicitly transparent surfaces. */
+	if(sd->flag & SD_HAS_ONLY_VOLUME) {
+		return;
+	}
+
 	float3 normal = make_float3(0.0f, 0.0f, 0.0f);
 	float3 albedo = make_float3(0.0f, 0.0f, 0.0f);
 	float sum_weight = 0.0f, sum_nonspecular_weight = 0.0f;
@@ -356,7 +361,16 @@ ccl_device_inline void kernel_write_result(KernelGlobals *kg, ccl_global float *
 #  endif
 			if(kernel_data.film.pass_denoising_clean) {
 				float3 noisy, clean;
-				path_radiance_split_denoising(kg, L, &noisy, &clean);
+#ifdef __SHADOW_TRICKS__
+				if(is_shadow_catcher) {
+					noisy = L_sum;
+					clean = make_float3(0.0f, 0.0f, 0.0f);
+				}
+				else
+#endif  /* __SHADOW_TRICKS__ */
+				{
+					path_radiance_split_denoising(kg, L, &noisy, &clean);
+				}
 				kernel_write_pass_float3_variance(buffer + kernel_data.film.pass_denoising_data + DENOISING_PASS_COLOR,
 				                                  sample, noisy);
 				kernel_write_pass_float3_unaligned(buffer + kernel_data.film.pass_denoising_clean,
@@ -364,7 +378,7 @@ ccl_device_inline void kernel_write_result(KernelGlobals *kg, ccl_global float *
 			}
 			else {
 				kernel_write_pass_float3_variance(buffer + kernel_data.film.pass_denoising_data + DENOISING_PASS_COLOR,
-				                                  sample, L_sum);
+				                                  sample, ensure_finite3(L_sum));
 			}
 
 			kernel_write_pass_float3_variance(buffer + kernel_data.film.pass_denoising_data + DENOISING_PASS_NORMAL,

intern/cycles/kernel/kernel_path.h

@@ -407,7 +407,7 @@ ccl_device void kernel_path_indirect(KernelGlobals *kg,
 		}
 #endif  /* __SUBSURFACE__ */
 
-#if defined(__EMISSION__) && defined(__BRANCHED_PATH__)
+#if defined(__EMISSION__)
 		if(kernel_data.integrator.use_direct_light) {
 			int all = (kernel_data.integrator.sample_all_lights_indirect) ||
 			          (state->flag & PATH_RAY_SHADOW_CATCHER);
@@ -421,7 +421,7 @@ ccl_device void kernel_path_indirect(KernelGlobals *kg,
 			                                           L,
 			                                           all);
 		}
-#endif  /* defined(__EMISSION__) && defined(__BRANCHED_PATH__) */
+#endif  /* defined(__EMISSION__) */
 
 		if(!kernel_path_surface_bounce(kg, rng, sd, &throughput, state, L, ray))
 			break;
@@ -643,11 +643,16 @@ ccl_device_inline float kernel_path_integrate(KernelGlobals *kg,
 #ifdef __SHADOW_TRICKS__
 		if((sd.object_flag & SD_OBJECT_SHADOW_CATCHER)) {
 			if(state.flag & PATH_RAY_CAMERA) {
-				state.flag |= (PATH_RAY_SHADOW_CATCHER | PATH_RAY_SHADOW_CATCHER_ONLY | PATH_RAY_STORE_SHADOW_INFO);
+				state.flag |= (PATH_RAY_SHADOW_CATCHER |
+				               PATH_RAY_SHADOW_CATCHER_ONLY |
+				               PATH_RAY_STORE_SHADOW_INFO);
 				state.catcher_object = sd.object;
 				if(!kernel_data.background.transparent) {
-					L->shadow_color = indirect_background(kg, &emission_sd, &state, &ray);
+					L->shadow_background_color =
+					        indirect_background(kg, &emission_sd, &state, &ray);
 				}
+				L->shadow_radiance_sum = path_radiance_clamp_and_sum(kg, L);
+				L->shadow_throughput = average(throughput);
 			}
 		}
 		else {

intern/cycles/kernel/kernel_path_branched.h

@@ -499,13 +499,16 @@ ccl_device float kernel_branched_path_integrate(KernelGlobals *kg,
 
 #ifdef __SHADOW_TRICKS__
 		if((sd.object_flag & SD_OBJECT_SHADOW_CATCHER)) {
-			if(state.flag & PATH_RAY_CAMERA) {
-				state.flag |= (PATH_RAY_SHADOW_CATCHER | PATH_RAY_SHADOW_CATCHER_ONLY | PATH_RAY_STORE_SHADOW_INFO);
-				state.catcher_object = sd.object;
-				if(!kernel_data.background.transparent) {
-					L->shadow_color = indirect_background(kg, &emission_sd, &state, &ray);
-				}
+			state.flag |= (PATH_RAY_SHADOW_CATCHER |
+			               PATH_RAY_SHADOW_CATCHER_ONLY |
+			               PATH_RAY_STORE_SHADOW_INFO);
+			state.catcher_object = sd.object;
+			if(!kernel_data.background.transparent) {
+				L->shadow_background_color =
+				        indirect_background(kg, &emission_sd, &state, &ray);
 			}
+			L->shadow_radiance_sum = path_radiance_clamp_and_sum(kg, L);
+			L->shadow_throughput = average(throughput);
 		}
 		else {
 			state.flag &= ~PATH_RAY_SHADOW_CATCHER_ONLY;

intern/cycles/kernel/kernel_path_state.h

@@ -139,9 +139,11 @@ ccl_device_inline void path_state_next(KernelGlobals *kg, ccl_addr_space PathSta
 	/* random number generator next bounce */
 	state->rng_offset += PRNG_BOUNCE_NUM;
 
+#ifdef __DENOISING_FEATURES__
 	if((state->denoising_feature_weight == 0.0f) && !(state->flag & PATH_RAY_SHADOW_CATCHER)) {
 		state->flag &= ~PATH_RAY_STORE_SHADOW_INFO;
 	}
+#endif
 }
 
 ccl_device_inline uint path_state_ray_visibility(KernelGlobals *kg, PathState *state)

intern/cycles/kernel/kernel_path_surface.h

@@ -16,7 +16,7 @@
 
 CCL_NAMESPACE_BEGIN
 
-#if defined(__BRANCHED_PATH__) || defined(__SUBSURFACE__) || defined(__SHADOW_TRICKS__)
+#if defined(__BRANCHED_PATH__) || defined(__SUBSURFACE__) || defined(__SHADOW_TRICKS__) || defined(__BAKING__)
 /* branched path tracing: connect path directly to position on one or more lights and add it to L */
 ccl_device_noinline void kernel_branched_path_surface_connect_light(
         KernelGlobals *kg,

intern/cycles/kernel/kernel_queues.h

@@ -128,6 +128,21 @@ ccl_device unsigned int get_global_queue_index(
 	return my_gqidx;
 }
 
+ccl_device int dequeue_ray_index(
+        int queue_number,
+        ccl_global int *queues,
+        int queue_size,
+        ccl_global int *queue_index)
+{
+	int index = atomic_fetch_and_dec_uint32((ccl_global uint*)&queue_index[queue_number])-1;
+
+	if(index < 0) {
+		return QUEUE_EMPTY_SLOT;
+	}
+
+	return queues[index + queue_number * queue_size];
+}
+
 CCL_NAMESPACE_END
 
 #endif // __KERNEL_QUEUE_H__

intern/cycles/kernel/kernel_subsurface.h

@@ -418,7 +418,7 @@ ccl_device_noinline void subsurface_scatter_multi_setup(
 }
 
 /* subsurface scattering step, from a point on the surface to another nearby point on the same object */
-ccl_device void subsurface_scatter_step(KernelGlobals *kg, ShaderData *sd, ccl_global PathState *state,
+ccl_device void subsurface_scatter_step(KernelGlobals *kg, ShaderData *sd, ccl_addr_space PathState *state,
 	int state_flag, ShaderClosure *sc, uint *lcg_state, float disk_u, float disk_v, bool all)
 {
 	float3 eval = make_float3(0.0f, 0.0f, 0.0f);
@@ -479,6 +479,10 @@ ccl_device void subsurface_scatter_step(KernelGlobals *kg, ShaderData *sd, ccl_g
 	if(ss_isect.num_hits > 0) {
 		float3 origP = sd->P;
 
+		/* Workaround for AMD GPU OpenCL compiler. Most probably cache bypass issue. */
+#if defined(__SPLIT_KERNEL__) && defined(__KERNEL_OPENCL_AMD__) && defined(__KERNEL_GPU__)
+		kernel_split_params.dummy_sd_flag = sd->flag;
+#endif
 		/* setup new shading point */
 		shader_setup_from_subsurface(kg, sd, &ss_isect.hits[0], &ray);
 

intern/cycles/kernel/kernel_textures.h

@@ -82,10 +82,10 @@ KERNEL_TEX(uint, texture_uint, __sobol_directions)
 #  if __CUDA_ARCH__ < 300
 /* full-float image */
 KERNEL_IMAGE_TEX(float4, texture_image_float4, __tex_image_float4_000)
-KERNEL_IMAGE_TEX(float4, texture_image_float4, __tex_image_float4_001)
-KERNEL_IMAGE_TEX(float4, texture_image_float4, __tex_image_float4_002)
-KERNEL_IMAGE_TEX(float4, texture_image_float4, __tex_image_float4_003)
-KERNEL_IMAGE_TEX(float4, texture_image_float4, __tex_image_float4_004)
+KERNEL_IMAGE_TEX(float4, texture_image_float4, __tex_image_float4_008)
+KERNEL_IMAGE_TEX(float4, texture_image_float4, __tex_image_float4_016)
+KERNEL_IMAGE_TEX(float4, texture_image_float4, __tex_image_float4_024)
+KERNEL_IMAGE_TEX(float4, texture_image_float4, __tex_image_float4_032)
 
 KERNEL_IMAGE_TEX(float4, texture_image3d_float4, __tex_image_float4_3d_000)
 KERNEL_IMAGE_TEX(float4, texture_image3d_float4, __tex_image_float4_3d_001)
@@ -93,91 +93,93 @@ KERNEL_IMAGE_TEX(float4, texture_image3d_float4, __tex_image_float4_3d_002)
 KERNEL_IMAGE_TEX(float4, texture_image3d_float4, __tex_image_float4_3d_003)
 KERNEL_IMAGE_TEX(float4, texture_image3d_float4, __tex_image_float4_3d_004)
 
-/* image */
-KERNEL_IMAGE_TEX(uchar4, texture_image_uchar4, __tex_image_byte4_005)
-KERNEL_IMAGE_TEX(uchar4, texture_image_uchar4, __tex_image_byte4_006)
-KERNEL_IMAGE_TEX(uchar4, texture_image_uchar4, __tex_image_byte4_007)
-KERNEL_IMAGE_TEX(uchar4, texture_image_uchar4, __tex_image_byte4_008)
+/* image
+ * These texture names are encoded to their flattened slots as
+ * ImageManager::type_index_to_flattened_slot() returns them. */
+KERNEL_IMAGE_TEX(uchar4, texture_image_uchar4, __tex_image_byte4_001)
 KERNEL_IMAGE_TEX(uchar4, texture_image_uchar4, __tex_image_byte4_009)
-KERNEL_IMAGE_TEX(uchar4, texture_image_uchar4, __tex_image_byte4_010)
-KERNEL_IMAGE_TEX(uchar4, texture_image_uchar4, __tex_image_byte4_011)
-KERNEL_IMAGE_TEX(uchar4, texture_image_uchar4, __tex_image_byte4_012)
-KERNEL_IMAGE_TEX(uchar4, texture_image_uchar4, __tex_image_byte4_013)
-KERNEL_IMAGE_TEX(uchar4, texture_image_uchar4, __tex_image_byte4_014)
-KERNEL_IMAGE_TEX(uchar4, texture_image_uchar4, __tex_image_byte4_015)
-KERNEL_IMAGE_TEX(uchar4, texture_image_uchar4, __tex_image_byte4_016)
 KERNEL_IMAGE_TEX(uchar4, texture_image_uchar4, __tex_image_byte4_017)
-KERNEL_IMAGE_TEX(uchar4, texture_image_uchar4, __tex_image_byte4_018)
-KERNEL_IMAGE_TEX(uchar4, texture_image_uchar4, __tex_image_byte4_019)
-KERNEL_IMAGE_TEX(uchar4, texture_image_uchar4, __tex_image_byte4_020)
-KERNEL_IMAGE_TEX(uchar4, texture_image_uchar4, __tex_image_byte4_021)
-KERNEL_IMAGE_TEX(uchar4, texture_image_uchar4, __tex_image_byte4_022)
-KERNEL_IMAGE_TEX(uchar4, texture_image_uchar4, __tex_image_byte4_023)
-KERNEL_IMAGE_TEX(uchar4, texture_image_uchar4, __tex_image_byte4_024)
 KERNEL_IMAGE_TEX(uchar4, texture_image_uchar4, __tex_image_byte4_025)
-KERNEL_IMAGE_TEX(uchar4, texture_image_uchar4, __tex_image_byte4_026)
-KERNEL_IMAGE_TEX(uchar4, texture_image_uchar4, __tex_image_byte4_027)
-KERNEL_IMAGE_TEX(uchar4, texture_image_uchar4, __tex_image_byte4_028)
-KERNEL_IMAGE_TEX(uchar4, texture_image_uchar4, __tex_image_byte4_029)
-KERNEL_IMAGE_TEX(uchar4, texture_image_uchar4, __tex_image_byte4_030)
-KERNEL_IMAGE_TEX(uchar4, texture_image_uchar4, __tex_image_byte4_031)
-KERNEL_IMAGE_TEX(uchar4, texture_image_uchar4, __tex_image_byte4_032)
 KERNEL_IMAGE_TEX(uchar4, texture_image_uchar4, __tex_image_byte4_033)
-KERNEL_IMAGE_TEX(uchar4, texture_image_uchar4, __tex_image_byte4_034)
-KERNEL_IMAGE_TEX(uchar4, texture_image_uchar4, __tex_image_byte4_035)
-KERNEL_IMAGE_TEX(uchar4, texture_image_uchar4, __tex_image_byte4_036)
-KERNEL_IMAGE_TEX(uchar4, texture_image_uchar4, __tex_image_byte4_037)
-KERNEL_IMAGE_TEX(uchar4, texture_image_uchar4, __tex_image_byte4_038)
-KERNEL_IMAGE_TEX(uchar4, texture_image_uchar4, __tex_image_byte4_039)
-KERNEL_IMAGE_TEX(uchar4, texture_image_uchar4, __tex_image_byte4_040)
 KERNEL_IMAGE_TEX(uchar4, texture_image_uchar4, __tex_image_byte4_041)
-KERNEL_IMAGE_TEX(uchar4, texture_image_uchar4, __tex_image_byte4_042)
-KERNEL_IMAGE_TEX(uchar4, texture_image_uchar4, __tex_image_byte4_043)
-KERNEL_IMAGE_TEX(uchar4, texture_image_uchar4, __tex_image_byte4_044)
-KERNEL_IMAGE_TEX(uchar4, texture_image_uchar4, __tex_image_byte4_045)
-KERNEL_IMAGE_TEX(uchar4, texture_image_uchar4, __tex_image_byte4_046)
-KERNEL_IMAGE_TEX(uchar4, texture_image_uchar4, __tex_image_byte4_047)
-KERNEL_IMAGE_TEX(uchar4, texture_image_uchar4, __tex_image_byte4_048)
 KERNEL_IMAGE_TEX(uchar4, texture_image_uchar4, __tex_image_byte4_049)
-KERNEL_IMAGE_TEX(uchar4, texture_image_uchar4, __tex_image_byte4_050)
-KERNEL_IMAGE_TEX(uchar4, texture_image_uchar4, __tex_image_byte4_051)
-KERNEL_IMAGE_TEX(uchar4, texture_image_uchar4, __tex_image_byte4_052)
-KERNEL_IMAGE_TEX(uchar4, texture_image_uchar4, __tex_image_byte4_053)
-KERNEL_IMAGE_TEX(uchar4, texture_image_uchar4, __tex_image_byte4_054)
-KERNEL_IMAGE_TEX(uchar4, texture_image_uchar4, __tex_image_byte4_055)
-KERNEL_IMAGE_TEX(uchar4, texture_image_uchar4, __tex_image_byte4_056)
 KERNEL_IMAGE_TEX(uchar4, texture_image_uchar4, __tex_image_byte4_057)
-KERNEL_IMAGE_TEX(uchar4, texture_image_uchar4, __tex_image_byte4_058)
-KERNEL_IMAGE_TEX(uchar4, texture_image_uchar4, __tex_image_byte4_059)
-KERNEL_IMAGE_TEX(uchar4, texture_image_uchar4, __tex_image_byte4_060)
-KERNEL_IMAGE_TEX(uchar4, texture_image_uchar4, __tex_image_byte4_061)
-KERNEL_IMAGE_TEX(uchar4, texture_image_uchar4, __tex_image_byte4_062)
-KERNEL_IMAGE_TEX(uchar4, texture_image_uchar4, __tex_image_byte4_063)
-KERNEL_IMAGE_TEX(uchar4, texture_image_uchar4, __tex_image_byte4_064)
 KERNEL_IMAGE_TEX(uchar4, texture_image_uchar4, __tex_image_byte4_065)
-KERNEL_IMAGE_TEX(uchar4, texture_image_uchar4, __tex_image_byte4_066)
-KERNEL_IMAGE_TEX(uchar4, texture_image_uchar4, __tex_image_byte4_067)
-KERNEL_IMAGE_TEX(uchar4, texture_image_uchar4, __tex_image_byte4_068)
-KERNEL_IMAGE_TEX(uchar4, texture_image_uchar4, __tex_image_byte4_069)
-KERNEL_IMAGE_TEX(uchar4, texture_image_uchar4, __tex_image_byte4_070)
-KERNEL_IMAGE_TEX(uchar4, texture_image_uchar4, __tex_image_byte4_071)
-KERNEL_IMAGE_TEX(uchar4, texture_image_uchar4, __tex_image_byte4_072)
 KERNEL_IMAGE_TEX(uchar4, texture_image_uchar4, __tex_image_byte4_073)
-KERNEL_IMAGE_TEX(uchar4, texture_image_uchar4, __tex_image_byte4_074)
-KERNEL_IMAGE_TEX(uchar4, texture_image_uchar4, __tex_image_byte4_075)
-KERNEL_IMAGE_TEX(uchar4, texture_image_uchar4, __tex_image_byte4_076)
-KERNEL_IMAGE_TEX(uchar4, texture_image_uchar4, __tex_image_byte4_077)
-KERNEL_IMAGE_TEX(uchar4, texture_image_uchar4, __tex_image_byte4_078)
-KERNEL_IMAGE_TEX(uchar4, texture_image_uchar4, __tex_image_byte4_079)
-KERNEL_IMAGE_TEX(uchar4, texture_image_uchar4, __tex_image_byte4_080)
 KERNEL_IMAGE_TEX(uchar4, texture_image_uchar4, __tex_image_byte4_081)
-KERNEL_IMAGE_TEX(uchar4, texture_image_uchar4, __tex_image_byte4_082)
-KERNEL_IMAGE_TEX(uchar4, texture_image_uchar4, __tex_image_byte4_083)
-KERNEL_IMAGE_TEX(uchar4, texture_image_uchar4, __tex_image_byte4_084)
-KERNEL_IMAGE_TEX(uchar4, texture_image_uchar4, __tex_image_byte4_085)
-KERNEL_IMAGE_TEX(uchar4, texture_image_uchar4, __tex_image_byte4_086)
-KERNEL_IMAGE_TEX(uchar4, texture_image_uchar4, __tex_image_byte4_087)
-KERNEL_IMAGE_TEX(uchar4, texture_image_uchar4, __tex_image_byte4_088)
+KERNEL_IMAGE_TEX(uchar4, texture_image_uchar4, __tex_image_byte4_089)
+KERNEL_IMAGE_TEX(uchar4, texture_image_uchar4, __tex_image_byte4_097)
+KERNEL_IMAGE_TEX(uchar4, texture_image_uchar4, __tex_image_byte4_105)
+KERNEL_IMAGE_TEX(uchar4, texture_image_uchar4, __tex_image_byte4_113)
+KERNEL_IMAGE_TEX(uchar4, texture_image_uchar4, __tex_image_byte4_121)
+KERNEL_IMAGE_TEX(uchar4, texture_image_uchar4, __tex_image_byte4_129)
+KERNEL_IMAGE_TEX(uchar4, texture_image_uchar4, __tex_image_byte4_137)
+KERNEL_IMAGE_TEX(uchar4, texture_image_uchar4, __tex_image_byte4_145)
+KERNEL_IMAGE_TEX(uchar4, texture_image_uchar4, __tex_image_byte4_153)
+KERNEL_IMAGE_TEX(uchar4, texture_image_uchar4, __tex_image_byte4_161)
+KERNEL_IMAGE_TEX(uchar4, texture_image_uchar4, __tex_image_byte4_169)
+KERNEL_IMAGE_TEX(uchar4, texture_image_uchar4, __tex_image_byte4_177)
+KERNEL_IMAGE_TEX(uchar4, texture_image_uchar4, __tex_image_byte4_185)
+KERNEL_IMAGE_TEX(uchar4, texture_image_uchar4, __tex_image_byte4_193)
+KERNEL_IMAGE_TEX(uchar4, texture_image_uchar4, __tex_image_byte4_201)
+KERNEL_IMAGE_TEX(uchar4, texture_image_uchar4, __tex_image_byte4_209)
+KERNEL_IMAGE_TEX(uchar4, texture_image_uchar4, __tex_image_byte4_217)
+KERNEL_IMAGE_TEX(uchar4, texture_image_uchar4, __tex_image_byte4_225)
+KERNEL_IMAGE_TEX(uchar4, texture_image_uchar4, __tex_image_byte4_233)
+KERNEL_IMAGE_TEX(uchar4, texture_image_uchar4, __tex_image_byte4_241)
+KERNEL_IMAGE_TEX(uchar4, texture_image_uchar4, __tex_image_byte4_249)
+KERNEL_IMAGE_TEX(uchar4, texture_image_uchar4, __tex_image_byte4_257)
+KERNEL_IMAGE_TEX(uchar4, texture_image_uchar4, __tex_image_byte4_265)
+KERNEL_IMAGE_TEX(uchar4, texture_image_uchar4, __tex_image_byte4_273)
+KERNEL_IMAGE_TEX(uchar4, texture_image_uchar4, __tex_image_byte4_281)
+KERNEL_IMAGE_TEX(uchar4, texture_image_uchar4, __tex_image_byte4_289)
+KERNEL_IMAGE_TEX(uchar4, texture_image_uchar4, __tex_image_byte4_297)
+KERNEL_IMAGE_TEX(uchar4, texture_image_uchar4, __tex_image_byte4_305)
+KERNEL_IMAGE_TEX(uchar4, texture_image_uchar4, __tex_image_byte4_313)
+KERNEL_IMAGE_TEX(uchar4, texture_image_uchar4, __tex_image_byte4_321)
+KERNEL_IMAGE_TEX(uchar4, texture_image_uchar4, __tex_image_byte4_329)
+KERNEL_IMAGE_TEX(uchar4, texture_image_uchar4, __tex_image_byte4_337)
+KERNEL_IMAGE_TEX(uchar4, texture_image_uchar4, __tex_image_byte4_345)
+KERNEL_IMAGE_TEX(uchar4, texture_image_uchar4, __tex_image_byte4_353)
+KERNEL_IMAGE_TEX(uchar4, texture_image_uchar4, __tex_image_byte4_361)
+KERNEL_IMAGE_TEX(uchar4, texture_image_uchar4, __tex_image_byte4_369)
+KERNEL_IMAGE_TEX(uchar4, texture_image_uchar4, __tex_image_byte4_377)
+KERNEL_IMAGE_TEX(uchar4, texture_image_uchar4, __tex_image_byte4_385)
+KERNEL_IMAGE_TEX(uchar4, texture_image_uchar4, __tex_image_byte4_393)
+KERNEL_IMAGE_TEX(uchar4, texture_image_uchar4, __tex_image_byte4_401)
+KERNEL_IMAGE_TEX(uchar4, texture_image_uchar4, __tex_image_byte4_409)
+KERNEL_IMAGE_TEX(uchar4, texture_image_uchar4, __tex_image_byte4_417)
+KERNEL_IMAGE_TEX(uchar4, texture_image_uchar4, __tex_image_byte4_425)
+KERNEL_IMAGE_TEX(uchar4, texture_image_uchar4, __tex_image_byte4_433)
+KERNEL_IMAGE_TEX(uchar4, texture_image_uchar4, __tex_image_byte4_441)
+KERNEL_IMAGE_TEX(uchar4, texture_image_uchar4, __tex_image_byte4_449)
+KERNEL_IMAGE_TEX(uchar4, texture_image_uchar4, __tex_image_byte4_457)
+KERNEL_IMAGE_TEX(uchar4, texture_image_uchar4, __tex_image_byte4_465)
+KERNEL_IMAGE_TEX(uchar4, texture_image_uchar4, __tex_image_byte4_473)
+KERNEL_IMAGE_TEX(uchar4, texture_image_uchar4, __tex_image_byte4_481)
+KERNEL_IMAGE_TEX(uchar4, texture_image_uchar4, __tex_image_byte4_489)
+KERNEL_IMAGE_TEX(uchar4, texture_image_uchar4, __tex_image_byte4_497)
+KERNEL_IMAGE_TEX(uchar4, texture_image_uchar4, __tex_image_byte4_505)
+KERNEL_IMAGE_TEX(uchar4, texture_image_uchar4, __tex_image_byte4_513)
+KERNEL_IMAGE_TEX(uchar4, texture_image_uchar4, __tex_image_byte4_521)
+KERNEL_IMAGE_TEX(uchar4, texture_image_uchar4, __tex_image_byte4_529)
+KERNEL_IMAGE_TEX(uchar4, texture_image_uchar4, __tex_image_byte4_537)
+KERNEL_IMAGE_TEX(uchar4, texture_image_uchar4, __tex_image_byte4_545)
+KERNEL_IMAGE_TEX(uchar4, texture_image_uchar4, __tex_image_byte4_553)
+KERNEL_IMAGE_TEX(uchar4, texture_image_uchar4, __tex_image_byte4_561)
+KERNEL_IMAGE_TEX(uchar4, texture_image_uchar4, __tex_image_byte4_569)
+KERNEL_IMAGE_TEX(uchar4, texture_image_uchar4, __tex_image_byte4_577)
+KERNEL_IMAGE_TEX(uchar4, texture_image_uchar4, __tex_image_byte4_585)
+KERNEL_IMAGE_TEX(uchar4, texture_image_uchar4, __tex_image_byte4_593)
+KERNEL_IMAGE_TEX(uchar4, texture_image_uchar4, __tex_image_byte4_601)
+KERNEL_IMAGE_TEX(uchar4, texture_image_uchar4, __tex_image_byte4_609)
+KERNEL_IMAGE_TEX(uchar4, texture_image_uchar4, __tex_image_byte4_617)
+KERNEL_IMAGE_TEX(uchar4, texture_image_uchar4, __tex_image_byte4_625)
+KERNEL_IMAGE_TEX(uchar4, texture_image_uchar4, __tex_image_byte4_633)
+KERNEL_IMAGE_TEX(uchar4, texture_image_uchar4, __tex_image_byte4_641)
+KERNEL_IMAGE_TEX(uchar4, texture_image_uchar4, __tex_image_byte4_649)
+KERNEL_IMAGE_TEX(uchar4, texture_image_uchar4, __tex_image_byte4_657)
+KERNEL_IMAGE_TEX(uchar4, texture_image_uchar4, __tex_image_byte4_665)
 
 #  else
 /* bindless textures */

intern/cycles/kernel/kernel_types.h

@@ -124,17 +124,19 @@ CCL_NAMESPACE_BEGIN
 #    define __VOLUME_SCATTER__
 #    define __SHADOW_RECORD_ALL__
 #    define __CMJ__
+#    define __BRANCHED_PATH__
 #  endif  /* __KERNEL_OPENCL_NVIDIA__ */
 
 #  ifdef __KERNEL_OPENCL_APPLE__
 #    define __KERNEL_SHADING__
 #    define __KERNEL_ADV_SHADING__
+#    define __PRINCIPLED__
 #    define __CMJ__
 /* TODO(sergey): Currently experimental section is ignored here,
  * this is because megakernel in device_opencl does not support
  * custom cflags depending on the scene features.
  */
-#  endif  /* __KERNEL_OPENCL_NVIDIA__ */
+#  endif  /* __KERNEL_OPENCL_APPLE__ */
 
 #  ifdef __KERNEL_OPENCL_AMD__
 #    define __CL_USE_NATIVE__
@@ -153,6 +155,7 @@ CCL_NAMESPACE_BEGIN
 #    define __CL_USE_NATIVE__
 #    define __KERNEL_SHADING__
 #    define __KERNEL_ADV_SHADING__
+#    define __PRINCIPLED__
 #    define __CMJ__
 #  endif  /* __KERNEL_OPENCL_INTEL_CPU__ */
 
@@ -235,6 +238,9 @@ CCL_NAMESPACE_BEGIN
 #ifdef __NO_PRINCIPLED__
 #  undef __PRINCIPLED__
 #endif
+#ifdef __NO_DENOISING__
+#  undef __DENOISING_FEATURES__
+#endif
 
 /* Random Numbers */
 
@@ -511,7 +517,13 @@ typedef ccl_addr_space struct PathRadiance {
 	float3 path_total_shaded;
 
 	/* Color of the background on which shadow is alpha-overed. */
-	float3 shadow_color;
+	float3 shadow_background_color;
+
+	/* Path radiance sum and throughput at the moment when ray hits shadow
+	 * catcher object.
+	 */
+	float3 shadow_radiance_sum;
+	float shadow_throughput;
 #endif
 
 #ifdef __DENOISING_FEATURES__
@@ -1386,6 +1398,8 @@ enum QueueNumber {
 #ifdef __BRANCHED_PATH__
 	/* All rays moving to next iteration of the indirect loop for light */
 	QUEUE_LIGHT_INDIRECT_ITER,
+	/* Queue of all inactive rays. These are candidates for sharing work of indirect loops */
+	QUEUE_INACTIVE_RAYS,
 #  ifdef __VOLUME__
 	/* All rays moving to next iteration of the indirect loop for volumes */
 	QUEUE_VOLUME_INDIRECT_ITER,
@@ -1428,6 +1442,9 @@ enum RayState {
 	RAY_BRANCHED_VOLUME_INDIRECT = (1 << 5),
 	RAY_BRANCHED_SUBSURFACE_INDIRECT = (1 << 6),
 	RAY_BRANCHED_INDIRECT = (RAY_BRANCHED_LIGHT_INDIRECT | RAY_BRANCHED_VOLUME_INDIRECT | RAY_BRANCHED_SUBSURFACE_INDIRECT),
+
+	/* Ray is evaluating an iteration of an indirect loop for another thread */
+	RAY_BRANCHED_INDIRECT_SHARED = (1 << 7),
 };
 
 #define ASSIGN_RAY_STATE(ray_state, ray_index, state) (ray_state[ray_index] = ((ray_state[ray_index] & RAY_FLAG_MASK) | state))

intern/cycles/kernel/kernel_volume.h

@@ -660,6 +660,7 @@ typedef struct VolumeSegment {
  * but the entire segment is needed to do always scattering, rather than probabilistically
  * hitting or missing the volume. if we don't know the transmittance at the end of the
  * volume we can't generate stratified distance samples up to that transmittance */
+#ifdef __VOLUME_DECOUPLED__
 ccl_device void kernel_volume_decoupled_record(KernelGlobals *kg, PathState *state,
 	Ray *ray, ShaderData *sd, VolumeSegment *segment, bool heterogeneous)
 {
@@ -829,6 +830,7 @@ ccl_device void kernel_volume_decoupled_free(KernelGlobals *kg, VolumeSegment *s
 #endif
 	}
 }
+#endif  /* __VOLUME_DECOUPLED__ */
 
 /* scattering for homogeneous and heterogeneous volumes, using decoupled ray
  * marching.

intern/cycles/kernel/kernels/cpu/filter.cpp

@@ -1,5 +1,5 @@
 /*
- * Copyright 2011-2013 Blender Foundation
+ * Copyright 2011-2017 Blender Foundation
  *
  * Licensed under the Apache License, Version 2.0 (the "License");
  * you may not use this file except in compliance with the License.