fix a minor bug when compiling

Was giving an issues in the past, will avoid it for now.

CMakeLists.txt

@@ -166,7 +166,7 @@ option_defaults_init(
 )
 
 # customize...
-if (UNIX AND NOT APPLE)
+if(UNIX AND NOT APPLE)
 	# some of these libraries are problematic on Linux
 	# disable less important dependencies by default
 	set(_init_CODEC_FFMPEG                   OFF)
@@ -218,6 +218,10 @@ if(${CMAKE_VERSION} VERSION_LESS 2.8.8)
 	# add_library OBJECT arg unsupported
 	set(WITH_BUILDINFO OFF)
 endif()
+set(BUILDINFO_OVERRIDE_DATE "" CACHE STRING "Use instead of the current date for reproducable builds (empty string disables this option)")
+set(BUILDINFO_OVERRIDE_TIME "" CACHE STRING "Use instead of the current time for reproducable builds (empty string disables this option)")
+mark_as_advanced(BUILDINFO_OVERRIDE_DATE)
+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)
@@ -362,19 +366,23 @@ if(UNIX AND NOT APPLE)
 		mark_as_advanced(WITH_BOOST_ICU)
 	endif()
 endif()
+
 option(WITH_PYTHON_INSTALL       "Copy system python into the blender install folder" ON)
-option(WITH_PYTHON_INSTALL_NUMPY "Copy system numpy into the blender install folder"  ON)
-set(PYTHON_NUMPY_PATH            "" CACHE PATH "Path to python site-packages or dist-packages containing 'numpy' module")
-mark_as_advanced(PYTHON_NUMPY_PATH)
+if(WITH_PYTHON_INSTALL)
+	option(WITH_PYTHON_INSTALL_NUMPY "Copy system numpy into the blender install folder"  ON)
+	set(PYTHON_NUMPY_PATH            "" CACHE PATH "Path to python site-packages or dist-packages containing 'numpy' module")
+	mark_as_advanced(PYTHON_NUMPY_PATH)
+
+	if(UNIX AND NOT APPLE)
+		option(WITH_PYTHON_INSTALL_REQUESTS "Copy system requests into the blender install folder" ON)
+		set(PYTHON_REQUESTS_PATH "" CACHE PATH "Path to python site-packages or dist-packages containing 'requests' module")
+		mark_as_advanced(PYTHON_REQUESTS_PATH)
+	endif()
+endif()
+
 option(WITH_CPU_SSE              "Enable SIMD instruction if they're detected on the host machine" ON)
 mark_as_advanced(WITH_CPU_SSE)
 
-if(UNIX AND NOT APPLE)
-	option(WITH_PYTHON_INSTALL_REQUESTS "Copy system requests into the blender install folder" ON)
-	set(PYTHON_REQUESTS_PATH "" CACHE PATH "Path to python site-packages or dist-packages containing 'requests' module")
-	mark_as_advanced(PYTHON_REQUESTS_PATH)
-endif()
-
 # Cycles
 option(WITH_CYCLES					"Enable Cycles Render Engine" ON)
 option(WITH_CYCLES_STANDALONE		"Build Cycles standalone application" OFF)
@@ -386,8 +394,10 @@ mark_as_advanced(CYCLES_CUDA_BINARIES_ARCH)
 unset(PLATFORM_DEFAULT)
 option(WITH_CYCLES_LOGGING	"Build Cycles with logging support" ON)
 option(WITH_CYCLES_DEBUG	"Build Cycles with extra debug capabilities" OFF)
+option(WITH_CYCLES_NATIVE_ONLY	"Build Cycles with native kernel only (which fits current CPU, use for development only)" OFF)
 mark_as_advanced(WITH_CYCLES_LOGGING)
 mark_as_advanced(WITH_CYCLES_DEBUG)
+mark_as_advanced(WITH_CYCLES_NATIVE_ONLY)
 
 option(WITH_CUDA_DYNLOAD "Dynamically load CUDA libraries at runtime" ON)
 mark_as_advanced(WITH_CUDA_DYNLOAD)
@@ -509,8 +519,8 @@ if(APPLE)
 
 	if(NOT CMAKE_OSX_ARCHITECTURES)
 		set(CMAKE_OSX_ARCHITECTURES x86_64 CACHE STRING
-		"Choose the architecture you want to build Blender for: i386, x86_64 or ppc"
-		FORCE)
+			"Choose the architecture you want to build Blender for: i386, x86_64 or ppc"
+			FORCE)
 	endif()
 
 	if(NOT DEFINED OSX_SYSTEM)
@@ -520,15 +530,20 @@ if(APPLE)
 		        OUTPUT_STRIP_TRAILING_WHITESPACE)
 	endif()
 
-	# workaround for incorrect cmake xcode lookup for developer previews - XCODE_VERSION does not take xcode-select path into accout
-	# but would always look into /Applications/Xcode.app while dev versions are named Xcode<version>-DP<preview_number>
-	execute_process(COMMAND xcode-select --print-path  OUTPUT_VARIABLE XCODE_CHECK OUTPUT_STRIP_TRAILING_WHITESPACE)
+	# workaround for incorrect cmake xcode lookup for developer previews - XCODE_VERSION does not
+	# take xcode-select path into account but would always look  into /Applications/Xcode.app
+	# while dev versions are named Xcode<version>-DP<preview_number>
+	execute_process(
+	        COMMAND xcode-select --print-path
+	        OUTPUT_VARIABLE XCODE_CHECK OUTPUT_STRIP_TRAILING_WHITESPACE)
 	string(REPLACE "/Contents/Developer" "" XCODE_BUNDLE ${XCODE_CHECK}) # truncate to bundlepath in any case
 	
 	if(${CMAKE_GENERATOR} MATCHES "Xcode")
 	
-		if(${XCODE_VERSION} VERSION_GREATER 4.2) # earlier xcode has no bundled developer dir, no sense in getting xcode path from
-			string(SUBSTRING "${XCODE_CHECK}" 14 6 DP_NAME) # reduce to XCode name without dp extension
+		# earlier xcode has no bundled developer dir, no sense in getting xcode path from
+		if(${XCODE_VERSION} VERSION_GREATER 4.2) 
+			# reduce to XCode name without dp extension
+			string(SUBSTRING "${XCODE_CHECK}" 14 6 DP_NAME) 
 			if(${DP_NAME} MATCHES Xcode5)
 				set(XCODE_VERSION 5)
 			endif()
@@ -555,25 +570,30 @@ if(APPLE)
 	message(STATUS "Detected OS X ${OSX_SYSTEM} and Xcode ${XCODE_VERSION} at ${XCODE_BUNDLE}")
 
 	if(${XCODE_VERSION} VERSION_LESS 4.3)
-		set(CMAKE_OSX_SYSROOT /Developer/SDKs/MacOSX${OSX_SYSTEM}.sdk CACHE PATH "" FORCE)  # use guaranteed existing sdk
+		# use guaranteed existing sdk
+		set(CMAKE_OSX_SYSROOT /Developer/SDKs/MacOSX${OSX_SYSTEM}.sdk CACHE PATH "" FORCE)
 	else()
-		# note: xcode-select path could be ambigous, cause /Applications/Xcode.app/Contents/Developer or /Applications/Xcode.app would be allowed
+		# note: xcode-select path could be ambigous,
+		# cause /Applications/Xcode.app/Contents/Developer or /Applications/Xcode.app would be allowed
 		# so i use a selfcomposed bundlepath here  
 		set(OSX_SYSROOT_PREFIX ${XCODE_BUNDLE}/Contents/Developer/Platforms/MacOSX.platform)
 		message(STATUS "OSX_SYSROOT_PREFIX: " ${OSX_SYSROOT_PREFIX})
 		set(OSX_DEVELOPER_PREFIX /Developer/SDKs/MacOSX${OSX_SYSTEM}.sdk) # use guaranteed existing sdk
 		set(CMAKE_OSX_SYSROOT ${OSX_SYSROOT_PREFIX}/${OSX_DEVELOPER_PREFIX} CACHE PATH "" FORCE)
 		if(${CMAKE_GENERATOR} MATCHES "Xcode")
-			set(CMAKE_XCODE_ATTRIBUTE_SDKROOT macosx${OSX_SYSTEM}) # to silence sdk not found warning, just overrides CMAKE_OSX_SYSROOT
+			# to silence sdk not found warning, just overrides CMAKE_OSX_SYSROOT
+			set(CMAKE_XCODE_ATTRIBUTE_SDKROOT macosx${OSX_SYSTEM})
 		endif()
 	endif()
 
 	if(OSX_SYSTEM MATCHES 10.9)
-		set(CMAKE_FIND_ROOT_PATH ${CMAKE_OSX_SYSROOT}) # make sure syslibs and headers are looked up in sdk ( expecially for 10.9 openGL atm. )
+		# make sure syslibs and headers are looked up in sdk ( expecially for 10.9 openGL atm. )
+		set(CMAKE_FIND_ROOT_PATH ${CMAKE_OSX_SYSROOT})
 	endif()
 
 	if(NOT CMAKE_OSX_DEPLOYMENT_TARGET)
-		set(CMAKE_OSX_DEPLOYMENT_TARGET "10.6" CACHE STRING "" FORCE) # 10.6 is our min. target, if you use higher sdk, weak linking happens
+		# 10.6 is our min. target, if you use higher sdk, weak linking happens
+		set(CMAKE_OSX_DEPLOYMENT_TARGET "10.6" CACHE STRING "" FORCE)
 	endif()
 	
 	if(NOT ${CMAKE_GENERATOR} MATCHES "Xcode")
@@ -582,8 +602,6 @@ if(APPLE)
 		set(CMAKE_CXX_FLAGS "${CMAKE_CXX_FLAGS} -mmacosx-version-min=${CMAKE_OSX_DEPLOYMENT_TARGET}")
 		add_definitions("-DMACOSX_DEPLOYMENT_TARGET=${CMAKE_OSX_DEPLOYMENT_TARGET}")
 	endif()
-
-	option(WITH_LIBS10.5  "Use 10.5 libs (needed for 64bit builds)" OFF)
 endif()
 
 
@@ -591,7 +609,10 @@ endif()
 # Check for conflicting/unsupported configurations
 
 if(NOT WITH_BLENDER AND NOT WITH_PLAYER AND NOT WITH_CYCLES_STANDALONE)
-	message(FATAL_ERROR "At least one of WITH_BLENDER or WITH_PLAYER or WITH_CYCLES_STANDALONE must be enabled, nothing to do!")
+	message(FATAL_ERROR
+		"At least one of WITH_BLENDER or WITH_PLAYER or "
+		"WITH_CYCLES_STANDALONE must be enabled, nothing to do!"
+	)
 endif()
 
 if(NOT WITH_GAMEENGINE AND WITH_PLAYER)
@@ -655,7 +676,8 @@ if(NOT WITH_BOOST)
 	set_and_warn(WITH_OPENAL         OFF)  # depends on AUDASPACE
 	set_and_warn(WITH_GAMEENGINE     OFF)  # depends on AUDASPACE
 	set_and_warn(WITH_PLAYER         OFF)  # depends on GAMEENGINE
-elseif(WITH_CYCLES OR WITH_OPENIMAGEIO OR WITH_AUDASPACE OR WITH_INTERNATIONAL OR WITH_OPENVDB OR WITH_OPENCOLORIO OR WITH_MOD_BOOLEAN)
+elseif(WITH_CYCLES OR WITH_OPENIMAGEIO OR WITH_AUDASPACE OR WITH_INTERNATIONAL OR
+       WITH_OPENVDB OR WITH_OPENCOLORIO OR WITH_MOD_BOOLEAN)
 	# Keep enabled
 else()
 	# New dependency graph needs either Boost or C++11 for function bindings.
@@ -752,18 +774,22 @@ endif()
 
 if(WITH_INTERNATIONAL)
 	if(NOT EXISTS "${CMAKE_SOURCE_DIR}/release/datafiles/locale/languages")
-		message(WARNING "Translation path '${CMAKE_SOURCE_DIR}/release/datafiles/locale' is missing, "
-						"This is a 'git submodule', which are known not to work with bridges to other version "
-						"control systems, disabling 'WITH_INTERNATIONAL'.")
+		message(WARNING
+			"Translation path '${CMAKE_SOURCE_DIR}/release/datafiles/locale' is missing, "
+			"This is a 'git submodule', which are known not to work with bridges to other version "
+			"control systems, disabling 'WITH_INTERNATIONAL'."
+		)
 		set(WITH_INTERNATIONAL OFF)
 	endif()
 endif()
 
 if(WITH_PYTHON)
 	if(NOT EXISTS "${CMAKE_SOURCE_DIR}/release/scripts/addons/modules")
-		message(WARNING "Addons path '${CMAKE_SOURCE_DIR}/release/scripts/addons' is missing, "
-						"This is a 'git submodule', which are known not to work with bridges to other version "
-						"control systems: * CONTINUING WITHOUT ADDONS *")
+		message(WARNING
+			"Addons path '${CMAKE_SOURCE_DIR}/release/scripts/addons' is missing, "
+			"This is a 'git submodule', which are known not to work with bridges to other version "
+			"control systems: * CONTINUING WITHOUT ADDONS *"
+		)
 	endif()
 endif()
 
@@ -797,21 +823,6 @@ set(PLATFORM_LINKFLAGS "")
 set(PLATFORM_LINKFLAGS_DEBUG "")
 
 
-# For alternate Python locations the commandline can be used to override detected/default cache settings, e.g:
-# On Unix:
-#   cmake ../blender \
-#         -D PYTHON_VERSION=3.5 \
-#         -D PYTHON_INCLUDE_DIR=/opt/py35/include/python3.5d \
-#         -D PYTHON_LIBRARY=/opt/py35/lib/libpython3.5d.so
-#
-# On Macs:
-#   cmake ../blender \
-#         -D PYTHON_INCLUDE_DIR=/System/Library/Frameworks/Python.framework/Versions/3.5/include/python3.5 \
-#         -D PYTHON_LIBPATH=/System/Library/Frameworks/Python.framework/Versions/3.5/lib/python3.5/config \
-#         -G Xcode
-#
-# When changing any of this remember to update the notes in doc/build_systems/cmake.txt
-
 #-----------------------------------------------------------------------------
 #Platform specifics
 
@@ -1034,7 +1045,11 @@ if(UNIX AND NOT APPLE)
 			if(${OSL_LIBRARY_VERSION_MAJOR} EQUAL "1" AND ${OSL_LIBRARY_VERSION_MINOR} LESS "6")
 				# Note: --whole-archive is needed to force loading of all symbols in liboslexec,
 				# otherwise LLVM is missing the osl_allocate_closure_component function
-				set(OSL_LIBRARIES ${OSL_OSLCOMP_LIBRARY} -Wl,--whole-archive ${OSL_OSLEXEC_LIBRARY} -Wl,--no-whole-archive ${OSL_OSLQUERY_LIBRARY})
+				set(OSL_LIBRARIES
+					${OSL_OSLCOMP_LIBRARY}
+					-Wl,--whole-archive ${OSL_OSLEXEC_LIBRARY}
+					-Wl,--no-whole-archive ${OSL_OSLQUERY_LIBRARY}
+				)
 			endif()
 		else()
 			message(STATUS "OSL not found, disabling it from Cycles")
@@ -1105,7 +1120,13 @@ if(UNIX AND NOT APPLE)
 			set(PUGIXML_LIBRARIES "")
 		endif()
 
-		set(OPENIMAGEIO_LIBRARIES ${OPENIMAGEIO_LIBRARIES} ${PNG_LIBRARIES} ${JPEG_LIBRARIES} ${ZLIB_LIBRARIES} ${BOOST_LIBRARIES})
+		set(OPENIMAGEIO_LIBRARIES
+			${OPENIMAGEIO_LIBRARIES}
+			${PNG_LIBRARIES}
+			${JPEG_LIBRARIES}
+			${ZLIB_LIBRARIES}
+			${BOOST_LIBRARIES}
+		)
 		set(OPENIMAGEIO_LIBPATH)  # TODO, remove and reference the absolute path everywhere
 		set(OPENIMAGEIO_DEFINITIONS "")
 
@@ -1146,7 +1167,9 @@ if(UNIX AND NOT APPLE)
 
 	if(WITH_LLVM OR WITH_SDL_DYNLOAD)
 		# Fix for conflict with Mesa llvmpipe
-		set(PLATFORM_LINKFLAGS "${PLATFORM_LINKFLAGS} -Wl,--version-script='${CMAKE_SOURCE_DIR}/source/creator/blender.map'")
+		set(PLATFORM_LINKFLAGS
+			"${PLATFORM_LINKFLAGS} -Wl,--version-script='${CMAKE_SOURCE_DIR}/source/creator/blender.map'"
+		)
 	endif()
 
 	if(WITH_OPENSUBDIV)
@@ -1249,7 +1272,10 @@ elseif(WIN32)
 		set(CMAKE_SHARED_LINKER_FLAGS "${CMAKE_SHARED_LINKER_FLAGS} /SAFESEH:NO")
 		set(CMAKE_MODULE_LINKER_FLAGS "${CMAKE_MODULE_LINKER_FLAGS} /SAFESEH:NO")
 
-		list(APPEND PLATFORM_LINKLIBS ws2_32 vfw32 winmm kernel32 user32 gdi32 comdlg32 advapi32 shfolder shell32 ole32 oleaut32 uuid psapi Dbghelp)
+		list(APPEND PLATFORM_LINKLIBS
+			ws2_32 vfw32 winmm kernel32 user32 gdi32 comdlg32
+			advapi32 shfolder shell32 ole32 oleaut32 uuid psapi Dbghelp
+		)
 
 		if(WITH_INPUT_IME)
 			list(APPEND PLATFORM_LINKLIBS imm32)
@@ -1287,7 +1313,8 @@ elseif(WIN32)
 		set(CMAKE_CXX_FLAGS_RELWITHDEBINFO "${CMAKE_CXX_FLAGS_RELWITHDEBINFO} /MT")
 		set(CMAKE_C_FLAGS_RELWITHDEBINFO "${CMAKE_C_FLAGS_RELWITHDEBINFO} /MT")
 
-		set(PLATFORM_LINKFLAGS "/SUBSYSTEM:CONSOLE /STACK:2097152 /INCREMENTAL:NO /NODEFAULTLIB:msvcrt.lib /NODEFAULTLIB:msvcmrt.lib /NODEFAULTLIB:msvcurt.lib /NODEFAULTLIB:msvcrtd.lib")
+		set(PLATFORM_LINKFLAGS "/SUBSYSTEM:CONSOLE /STACK:2097152 /INCREMENTAL:NO ")
+		set(PLATFORM_LINKFLAGS "${PLATFORM_LINKFLAGS} /NODEFAULTLIB:msvcrt.lib /NODEFAULTLIB:msvcmrt.lib /NODEFAULTLIB:msvcurt.lib /NODEFAULTLIB:msvcrtd.lib ")
 
 		# Ignore meaningless for us linker warnings.
 		set(PLATFORM_LINKFLAGS "${PLATFORM_LINKFLAGS} /ignore:4049 /ignore:4217 /ignore:4221")
@@ -1302,6 +1329,17 @@ elseif(WIN32)
 
 		set(PLATFORM_LINKFLAGS_DEBUG "/IGNORE:4099 /NODEFAULTLIB:libcmt.lib /NODEFAULTLIB:libc.lib")
 
+		# Use dynamic loading for OpenMP
+		if(WITH_OPENMP)
+			if(MSVC_VERSION EQUAL 1800)
+				set(OPENMP_DLL_NAME "vcomp120")
+			else()
+				set(OPENMP_DLL_NAME "vcomp140")
+			endif()
+			set(PLATFORM_LINKFLAGS "${PLATFORM_LINKFLAGS} /DELAYLOAD:${OPENMP_DLL_NAME}.dll delayimp.lib")
+			set(PLATFORM_LINKFLAGS_DEBUG "${PLATFORM_LINKFLAGS_DEBUG} /DELAYLOAD:${OPENMP_DLL_NAME}d.dll delayimp.lib")
+		endif()
+
 		if(NOT DEFINED LIBDIR)
 
 			# Setup 64bit and 64bit windows systems
@@ -1330,9 +1368,9 @@ elseif(WIN32)
 		# Add each of our libraries to our cmake_prefix_path so find_package() could work
 		file(GLOB children RELATIVE ${LIBDIR} ${LIBDIR}/*)
 		foreach(child ${children})
-		if(IS_DIRECTORY ${LIBDIR}/${child})
-			list(APPEND CMAKE_PREFIX_PATH  ${LIBDIR}/${child})
-		endif()
+			if(IS_DIRECTORY ${LIBDIR}/${child})
+				list(APPEND CMAKE_PREFIX_PATH  ${LIBDIR}/${child})
+			endif()
 		endforeach()
 
 		set(ZLIB_INCLUDE_DIRS ${LIBDIR}/zlib/include)
@@ -1437,23 +1475,28 @@ elseif(WIN32)
 				set(OPENEXR_INCLUDE_DIRS ${OPENEXR_INCLUDE_DIR} ${OPENEXR}/include/OpenEXR)
 				set(OPENEXR_LIBPATH ${OPENEXR}/lib)
 				set(OPENEXR_LIBRARIES
-					${OPENEXR_LIBPATH}/Iex-2_2.lib
-					${OPENEXR_LIBPATH}/Half.lib
-					${OPENEXR_LIBPATH}/IlmImf-2_2.lib
-					${OPENEXR_LIBPATH}/Imath-2_2.lib
-					${OPENEXR_LIBPATH}/IlmThread-2_2.lib
+					optimized ${OPENEXR_LIBPATH}/Iex-2_2.lib
+					optimized ${OPENEXR_LIBPATH}/Half.lib
+					optimized ${OPENEXR_LIBPATH}/IlmImf-2_2.lib
+					optimized ${OPENEXR_LIBPATH}/Imath-2_2.lib
+					optimized ${OPENEXR_LIBPATH}/IlmThread-2_2.lib
+					debug ${OPENEXR_LIBPATH}/Iex-2_2_d.lib
+					debug ${OPENEXR_LIBPATH}/Half_d.lib
+					debug ${OPENEXR_LIBPATH}/IlmImf-2_2_d.lib
+					debug ${OPENEXR_LIBPATH}/Imath-2_2_d.lib
+					debug ${OPENEXR_LIBPATH}/IlmThread-2_2_d.lib
 				)
 			endif()
 		endif()
 
 		if(WITH_IMAGE_TIFF)
-		# Try to find tiff first then complain and set static and maybe wrong paths
-		find_package(TIFF)
-		if(NOT TIFF_FOUND)
-			message(WARNING "Using HARDCODED libtiff locations")
-			set(TIFF_LIBRARY ${LIBDIR}/tiff/lib/libtiff.lib)
-			set(TIFF_INCLUDE_DIR ${LIBDIR}/tiff/include)
-		endif()
+			# Try to find tiff first then complain and set static and maybe wrong paths
+			find_package(TIFF)
+			if(NOT TIFF_FOUND)
+				message(WARNING "Using HARDCODED libtiff locations")
+				set(TIFF_LIBRARY ${LIBDIR}/tiff/lib/libtiff.lib)
+				set(TIFF_INCLUDE_DIR ${LIBDIR}/tiff/include)
+			endif()
 		endif()
 
 		if(WITH_JACK)
@@ -1508,12 +1551,17 @@ elseif(WIN32)
 					set(BOOST_DEBUG_POSTFIX "vc140-mt-sgd-1_60.lib")
 				endif()
 				set(BOOST_LIBRARIES
-					optimized libboost_date_time-${BOOST_POSTFIX} optimized libboost_filesystem-${BOOST_POSTFIX}
+					optimized libboost_date_time-${BOOST_POSTFIX}
+					optimized libboost_filesystem-${BOOST_POSTFIX}
 					optimized libboost_regex-${BOOST_POSTFIX}
-					optimized libboost_system-${BOOST_POSTFIX} optimized libboost_thread-${BOOST_POSTFIX}
-					debug libboost_date_time-${BOOST_DEBUG_POSTFIX} debug libboost_filesystem-${BOOST_DEBUG_POSTFIX}
+					optimized libboost_system-${BOOST_POSTFIX}
+					optimized libboost_thread-${BOOST_POSTFIX}
+					debug libboost_date_time-${BOOST_DEBUG_POSTFIX}
+					debug libboost_filesystem-${BOOST_DEBUG_POSTFIX}
 					debug libboost_regex-${BOOST_DEBUG_POSTFIX}
-					debug libboost_system-${BOOST_DEBUG_POSTFIX} debug libboost_thread-${BOOST_DEBUG_POSTFIX})
+					debug libboost_system-${BOOST_DEBUG_POSTFIX}
+					debug libboost_thread-${BOOST_DEBUG_POSTFIX}
+				)
 				if(WITH_CYCLES_OSL)
 					set(BOOST_LIBRARIES ${BOOST_LIBRARIES}
 						optimized libboost_wave-${BOOST_POSTFIX}
@@ -1580,12 +1628,12 @@ elseif(WIN32)
 		endif()
 
 		if(WITH_OPENVDB)
-			set(BLOSC_LIBRARIES ${LIBDIR}/blosc/lib/libblosc.lib)
-			set(TBB_LIBRARIES ${LIBDIR}/tbb/lib/tbb.lib)
+			set(BLOSC_LIBRARIES optimized ${LIBDIR}/blosc/lib/libblosc.lib debug ${LIBDIR}/blosc/lib/libblosc_d.lib)
+			set(TBB_LIBRARIES optimized ${LIBDIR}/tbb/lib/tbb.lib debug ${LIBDIR}/tbb/lib/tbb_debug.lib)
 			set(TBB_INCLUDE_DIR ${LIBDIR}/tbb/include)
 			set(OPENVDB ${LIBDIR}/openvdb)
 			set(OPENVDB_INCLUDE_DIRS ${OPENVDB}/include ${TBB_INCLUDE_DIR})
-			set(OPENVDB_LIBRARIES openvdb ${TBB_LIBRARIES} ${BLOSC_LIBRARIES})
+			set(OPENVDB_LIBRARIES optimized openvdb debug openvdb_d ${TBB_LIBRARIES} ${BLOSC_LIBRARIES})
 			set(OPENVDB_LIBPATH ${LIBDIR}/openvdb/lib)
 		endif()
 
@@ -1642,7 +1690,8 @@ elseif(WIN32)
 			#endif
 			int main(void) { return 0; }
 			" 
-			WITH_MINGW64)
+			WITH_MINGW64
+		)
 		
 		if(NOT DEFINED LIBDIR)
 			if(WITH_MINGW64)
@@ -1663,7 +1712,10 @@ elseif(WIN32)
 			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)
+		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)
@@ -1732,7 +1784,14 @@ elseif(WIN32)
 				${OPENCOLLADA}/include/opencollada/GeneratedSaxParser
 			)
 			set(OPENCOLLADA_LIBPATH ${OPENCOLLADA}/lib/opencollada)
-			set(OPENCOLLADA_LIBRARIES OpenCOLLADAStreamWriter OpenCOLLADASaxFrameworkLoader OpenCOLLADAFramework OpenCOLLADABaseUtils GeneratedSaxParser UTF MathMLSolver buffer ftoa xml)
+			set(OPENCOLLADA_LIBRARIES
+				OpenCOLLADAStreamWriter
+				OpenCOLLADASaxFrameworkLoader
+				OpenCOLLADAFramework
+				OpenCOLLADABaseUtils
+				GeneratedSaxParser
+				UTF MathMLSolver buffer ftoa xml
+			)
 			set(PCRE_LIBRARIES pcre)
 		endif()
 
@@ -1805,12 +1864,14 @@ elseif(WIN32)
 			if(WITH_INTERNATIONAL)
 				set(BOOST_LIBRARIES ${BOOST_LIBRARIES}
 					optimized boost_locale-${BOOST_POSTFIX}
-					debug boost_locale-${BOOST_DEBUG_POSTFIX}) 
+					debug boost_locale-${BOOST_DEBUG_POSTFIX}
+				)
 			endif()
 			if(WITH_CYCLES_OSL)
 				set(BOOST_LIBRARIES ${BOOST_LIBRARIES}
 					optimized boost_wave-${BOOST_POSTFIX}
-					debug boost_wave-${BOOST_DEBUG_POSTFIX}) 
+					debug boost_wave-${BOOST_DEBUG_POSTFIX}
+				)
 			endif()
 			set(BOOST_LIBPATH ${BOOST}/lib)
 			set(BOOST_DEFINITIONS "-DBOOST_ALL_NO_LIB -DBOOST_THREAD_USE_LIB ")
@@ -1919,7 +1980,14 @@ elseif(WIN32)
 		find_library(OSL_LIB_EXEC_DEBUG NAMES oslexec_d PATHS ${CYCLES_OSL}/lib)
 		find_library(OSL_LIB_COMP_DEBUG NAMES oslcomp_d PATHS ${CYCLES_OSL}/lib)
 		find_library(OSL_LIB_QUERY_DEBUG NAMES oslquery_d PATHS ${CYCLES_OSL}/lib)
-		list(APPEND OSL_LIBRARIES optimized ${OSL_LIB_COMP} optimized ${OSL_LIB_EXEC} optimized ${OSL_LIB_QUERY} debug ${OSL_LIB_EXEC_DEBUG} debug ${OSL_LIB_COMP_DEBUG} debug ${OSL_LIB_QUERY_DEBUG})
+		list(APPEND OSL_LIBRARIES
+			optimized ${OSL_LIB_COMP}
+			optimized ${OSL_LIB_EXEC}
+			optimized ${OSL_LIB_QUERY}
+			debug ${OSL_LIB_EXEC_DEBUG}
+			debug ${OSL_LIB_COMP_DEBUG}
+			debug ${OSL_LIB_QUERY_DEBUG}
+		)
 		find_path(OSL_INCLUDE_DIR OSL/oslclosure.h PATHS ${CYCLES_OSL}/include)
 		find_program(OSL_COMPILER NAMES oslc PATHS ${CYCLES_OSL}/bin)
 	
@@ -1933,20 +2001,8 @@ elseif(WIN32)
 
 elseif(APPLE)
 
-	if(${CMAKE_OSX_DEPLOYMENT_TARGET} STREQUAL "10.5" OR ${CMAKE_OSX_DEPLOYMENT_TARGET} STRGREATER "10.5")
-		set(WITH_LIBS10.5 ON CACHE BOOL "Use 10.5 libs" FORCE) # valid also for 10.6/7/8/9
-	endif()
-
 	if(NOT DEFINED LIBDIR)
-		if(WITH_LIBS10.5)
-			set(LIBDIR ${CMAKE_SOURCE_DIR}/../lib/darwin-9.x.universal)
-		else()
-			if(CMAKE_OSX_ARCHITECTURES MATCHES i386)
-				set(LIBDIR ${CMAKE_SOURCE_DIR}/../lib/darwin-8.x.i386)
-			else()
-				set(LIBDIR ${CMAKE_SOURCE_DIR}/../lib/darwin-8.0.0-powerpc)
-			endif()
-		endif()
+		set(LIBDIR ${CMAKE_SOURCE_DIR}/../lib/darwin-9.x.universal)
 	else()
 		message(STATUS "Using pre-compiled LIBDIR: ${LIBDIR}")
 	endif()
@@ -1998,22 +2054,30 @@ elseif(APPLE)
 		if(NOT WITH_PYTHON_MODULE AND NOT WITH_PYTHON_FRAMEWORK)
 			# normally cached but not since we include them with blender
 			set(PYTHON_INCLUDE_DIR "${LIBDIR}/python/include/python${PYTHON_VERSION}m")
-			# set(PYTHON_EXECUTABLE "${LIBDIR}/python/bin/python${PYTHON_VERSION}") # not used yet
+			set(PYTHON_EXECUTABLE "${LIBDIR}/python/bin/python${PYTHON_VERSION}m")
 			set(PYTHON_LIBRARY python${PYTHON_VERSION}m)
 			set(PYTHON_LIBPATH "${LIBDIR}/python/lib/python${PYTHON_VERSION}")
 			# set(PYTHON_LINKFLAGS "-u _PyMac_Error")  # won't  build with this enabled
 		else()
 			# module must be compiled against Python framework
-			set(PYTHON_INCLUDE_DIR "/Library/Frameworks/Python.framework/Versions/${PYTHON_VERSION}/include/python${PYTHON_VERSION}m")
-			set(PYTHON_EXECUTABLE "/Library/Frameworks/Python.framework/Versions/${PYTHON_VERSION}/bin/python${PYTHON_VERSION}")
+			set(_py_framework "/Library/Frameworks/Python.framework/Versions/${PYTHON_VERSION}")
+
+			set(PYTHON_INCLUDE_DIR "${_py_framework}/include/python${PYTHON_VERSION}m")
+			set(PYTHON_EXECUTABLE "${_py_framework}/bin/python${PYTHON_VERSION}m")
+			set(PYTHON_LIBPATH "${_py_framework}/lib/python${PYTHON_VERSION}/config-${PYTHON_VERSION}m")
 			#set(PYTHON_LIBRARY python${PYTHON_VERSION})
-			set(PYTHON_LIBPATH "/Library/Frameworks/Python.framework/Versions/${PYTHON_VERSION}/lib/python${PYTHON_VERSION}/config-${PYTHON_VERSION}m")
 			#set(PYTHON_LINKFLAGS "-u _PyMac_Error -framework Python")  # won't  build with this enabled
+
+			unset(_py_framework)
 		endif()
 		
 		# uncached vars
 		set(PYTHON_INCLUDE_DIRS "${PYTHON_INCLUDE_DIR}")
 		set(PYTHON_LIBRARIES  "${PYTHON_LIBRARY}")
+
+		if(NOT EXISTS "${PYTHON_EXECUTABLE}")
+			message(FATAL_ERROR "Python executable missing: ${PYTHON_EXECUTABLE}")
+		endif()
 	endif()
 
 	if(WITH_FFTW3)
@@ -2046,7 +2110,10 @@ elseif(APPLE)
 	if(WITH_CODEC_FFMPEG)
 		set(FFMPEG ${LIBDIR}/ffmpeg)
 		set(FFMPEG_INCLUDE_DIRS ${FFMPEG}/include)
-		set(FFMPEG_LIBRARIES avcodec avdevice avformat avutil mp3lame swscale x264 xvidcore theora theoradec theoraenc vorbis vorbisenc vorbisfile ogg)
+		set(FFMPEG_LIBRARIES
+			avcodec avdevice avformat avutil
+			mp3lame swscale x264 xvidcore theora theoradec theoraenc vorbis vorbisenc vorbisfile ogg
+		)
 		set(FFMPEG_LIBPATH ${FFMPEG}/lib)
 	endif()
 
@@ -2057,13 +2124,13 @@ elseif(APPLE)
 	)
 	mark_as_advanced(SYSTEMSTUBS_LIBRARY)
 	if(SYSTEMSTUBS_LIBRARY)
-		list(APPEND PLATFORM_LINKLIBS stdc++ SystemStubs)
-	else()
-		list(APPEND PLATFORM_LINKLIBS stdc++)
+		list(APPEND PLATFORM_LINKLIBS SystemStubs)
 	endif()
 
 	set(PLATFORM_CFLAGS "-pipe -funsigned-char")
-	set(PLATFORM_LINKFLAGS "-fexceptions -framework CoreServices -framework Foundation -framework IOKit -framework AppKit -framework Cocoa -framework Carbon -framework AudioUnit -framework AudioToolbox -framework CoreAudio")
+	set(PLATFORM_LINKFLAGS
+		"-fexceptions -framework CoreServices -framework Foundation -framework IOKit -framework AppKit -framework Cocoa -framework Carbon -framework AudioUnit -framework AudioToolbox -framework CoreAudio"
+	)
 	if(WITH_CODEC_QUICKTIME)
 		set(PLATFORM_LINKFLAGS "${PLATFORM_LINKFLAGS} -framework QTKit")
 		if(CMAKE_OSX_ARCHITECTURES MATCHES i386)
@@ -2072,6 +2139,12 @@ elseif(APPLE)
 		endif()
 	endif()
 
+	if(WITH_CXX11)
+		list(APPEND PLATFORM_LINKLIBS c++)
+	else()
+		list(APPEND PLATFORM_LINKLIBS stdc++)
+	endif()
+
 	# XXX - SOME MAC DEV PLEASE TEST WITH THE SDK INSTALLED!
 	# ALSO SHOULD BE MOVED INTO OWN MODULE WHEN FUNCTIONAL
 	if(WITH_INPUT_NDOF)
@@ -2095,7 +2168,8 @@ elseif(APPLE)
 	endif()
 	
 	if(WITH_PYTHON_MODULE OR WITH_PYTHON_FRAMEWORK)
-		set(PLATFORM_LINKFLAGS "${PLATFORM_LINKFLAGS} /Library/Frameworks/Python.framework/Versions/${PYTHON_VERSION}/Python")# force cmake to link right framework
+		# force cmake to link right framework
+		set(PLATFORM_LINKFLAGS "${PLATFORM_LINKFLAGS} /Library/Frameworks/Python.framework/Versions/${PYTHON_VERSION}/Python")
 	endif()
 	
 	if(WITH_OPENCOLLADA)
@@ -2110,10 +2184,18 @@ elseif(APPLE)
 		)
 
 		set(OPENCOLLADA_LIBPATH ${OPENCOLLADA}/lib)
-		set(OPENCOLLADA_LIBRARIES "OpenCOLLADASaxFrameworkLoader -lOpenCOLLADAFramework -lOpenCOLLADABaseUtils -lOpenCOLLADAStreamWriter -lMathMLSolver -lGeneratedSaxParser -lxml2 -lbuffer -lftoa")
+		set(OPENCOLLADA_LIBRARIES
+			OpenCOLLADASaxFrameworkLoader
+			-lOpenCOLLADAFramework
+			-lOpenCOLLADABaseUtils
+			-lOpenCOLLADAStreamWriter
+			-lMathMLSolver
+			-lGeneratedSaxParser
+			-lxml2 -lbuffer -lftoa
+		)
 		# Use UTF functions from collada if LLVM is not enabled
 		if(NOT WITH_LLVM)
-			set(OPENCOLLADA_LIBRARIES "${OPENCOLLADA_LIBRARIES} -lUTF")
+			list(APPEND OPENCOLLADA_LIBRARIES -lUTF)
 		endif()
 		# pcre is bundled with openCollada
 		#set(PCRE ${LIBDIR}/pcre)
@@ -2155,7 +2237,14 @@ elseif(APPLE)
 	if(WITH_BOOST)
 		set(BOOST ${LIBDIR}/boost)
 		set(BOOST_INCLUDE_DIR ${BOOST}/include)
-		set(BOOST_LIBRARIES boost_date_time-mt boost_filesystem-mt boost_regex-mt boost_system-mt boost_thread-mt boost_wave-mt)
+		set(BOOST_LIBRARIES
+			boost_date_time-mt
+			boost_filesystem-mt
+			boost_regex-mt
+			boost_system-mt
+			boost_thread-mt
+			boost_wave-mt
+		)
 		if(WITH_INTERNATIONAL)
 			list(APPEND BOOST_LIBRARIES boost_locale-mt)
 		endif()
@@ -2176,8 +2265,22 @@ elseif(APPLE)
 	if(WITH_OPENIMAGEIO)
 		set(OPENIMAGEIO ${LIBDIR}/openimageio)
 		set(OPENIMAGEIO_INCLUDE_DIRS ${OPENIMAGEIO}/include)
-		set(OPENIMAGEIO_LIBRARIES ${OPENIMAGEIO}/lib/libOpenImageIO.a ${PNG_LIBRARIES} ${JPEG_LIBRARIES} ${TIFF_LIBRARY} ${OPENEXR_LIBRARIES} ${ZLIB_LIBRARIES})
-		set(OPENIMAGEIO_LIBPATH ${OPENIMAGEIO}/lib ${JPEG_LIBPATH} ${PNG_LIBPATH} ${TIFF_LIBPATH} ${OPENEXR_LIBPATH} ${ZLIB_LIBPATH})
+		set(OPENIMAGEIO_LIBRARIES
+			${OPENIMAGEIO}/lib/libOpenImageIO.a
+			${PNG_LIBRARIES}
+			${JPEG_LIBRARIES}
+			${TIFF_LIBRARY}
+			${OPENEXR_LIBRARIES}
+			${ZLIB_LIBRARIES}
+		)
+		set(OPENIMAGEIO_LIBPATH
+			${OPENIMAGEIO}/lib
+			${JPEG_LIBPATH}
+			${PNG_LIBPATH}
+			${TIFF_LIBPATH}
+			${OPENEXR_LIBPATH}
+			${ZLIB_LIBPATH}
+		)
 		set(OPENIMAGEIO_DEFINITIONS "-DOIIO_STATIC_BUILD")
 		set(OPENIMAGEIO_IDIFF "${LIBDIR}/openimageio/bin/idiff")
 	endif()
@@ -2270,8 +2373,14 @@ elseif(APPLE)
 				include_directories(${LIBDIR}/openmp/include)
 				link_directories(${LIBDIR}/openmp/lib)
 				# This is a workaround for our helperbinaries ( datatoc, masgfmt, ... ),
-				# They are linked also to omp lib, so we need it in builddir for runtime exexcution, TODO: remove all unneeded dependencies from these
-				execute_process(COMMAND ditto -arch ${CMAKE_OSX_ARCHITECTURES} ${LIBDIR}/openmp/lib/libiomp5.dylib ${CMAKE_BINARY_DIR}/Resources/lib/libiomp5.dylib) # for intermediate binaries, in respect to lib ID
+				# They are linked also to omp lib, so we need it in builddir for runtime exexcution,
+				# TODO: remove all unneeded dependencies from these
+
+				# for intermediate binaries, in respect to lib ID
+				execute_process(
+				        COMMAND ditto -arch ${CMAKE_OSX_ARCHITECTURES}
+				        ${LIBDIR}/openmp/lib/libiomp5.dylib
+				        ${CMAKE_BINARY_DIR}/Resources/lib/libiomp5.dylib)
 			endif()
 		endif()
 	endif()
@@ -2297,7 +2406,20 @@ elseif(APPLE)
 		set(CMAKE_CXX_FLAGS "${CMAKE_CXX_FLAGS} -ftemplate-depth=1024")
 	endif()
 	# Get rid of eventually clashes, we export some symbols explicite as local
-	set(PLATFORM_LINKFLAGS "${PLATFORM_LINKFLAGS} -Xlinker -unexported_symbols_list -Xlinker ${CMAKE_SOURCE_DIR}/source/creator/osx_locals.map")
+	set(PLATFORM_LINKFLAGS
+		"${PLATFORM_LINKFLAGS} -Xlinker -unexported_symbols_list -Xlinker ${CMAKE_SOURCE_DIR}/source/creator/osx_locals.map"
+	)
+
+	if(WITH_CXX11)
+		set(CMAKE_CXX_FLAGS "${CMAKE_CXX_FLAGS} -stdlib=libc++")
+		set(PLATFORM_LINKFLAGS "${PLATFORM_LINKFLAGS} -stdlib=libc++")
+	endif()
+
+	# Suppress ranlib "has no symbols" warnings (workaround for T48250)
+	set(CMAKE_C_ARCHIVE_CREATE   "<CMAKE_AR> Scr <TARGET> <LINK_FLAGS> <OBJECTS>")
+	set(CMAKE_CXX_ARCHIVE_CREATE "<CMAKE_AR> Scr <TARGET> <LINK_FLAGS> <OBJECTS>")
+	set(CMAKE_C_ARCHIVE_FINISH   "<CMAKE_RANLIB> -no_warning_for_no_symbols -c <TARGET>")
+	set(CMAKE_CXX_ARCHIVE_FINISH "<CMAKE_RANLIB> -no_warning_for_no_symbols -c <TARGET>")
 endif()
 
 #-----------------------------------------------------------------------------
@@ -2309,38 +2431,50 @@ endif()
 
 if(WITH_CYCLES)
 	if(NOT WITH_OPENIMAGEIO)
-		message(FATAL_ERROR "Cycles requires WITH_OPENIMAGEIO, the library may not have been found. Configure OIIO or disable WITH_CYCLES")
+		message(FATAL_ERROR
+			"Cycles requires WITH_OPENIMAGEIO, the library may not have been found. "
+			"Configure OIIO or disable WITH_CYCLES"
+		)
 	endif()
 	if(NOT WITH_BOOST)
-		message(FATAL_ERROR "Cycles requires WITH_BOOST, the library may not have been found. Configure BOOST or disable WITH_CYCLES")
+		message(FATAL_ERROR
+			"Cycles requires WITH_BOOST, the library may not have been found. "
+			"Configure BOOST or disable WITH_CYCLES"
+		)
 	endif()
 
 	if(WITH_CYCLES_OSL)
 		if(NOT WITH_LLVM)
-			message(FATAL_ERROR "Cycles OSL requires WITH_LLVM, the library may not have been found. Configure LLVM or disable WITH_CYCLES_OSL")
+			message(FATAL_ERROR
+				"Cycles OSL requires WITH_LLVM, the library may not have been found. "
+				"Configure LLVM or disable WITH_CYCLES_OSL"
+			)
 		endif()
 	endif()
 endif()
 
 if(WITH_INTERNATIONAL)
 	if(NOT WITH_BOOST)
-		message(FATAL_ERROR "Internationalization requires WITH_BOOST, the library may not have been found. Configure BOOST or disable WITH_INTERNATIONAL")
+		message(FATAL_ERROR
+			"Internationalization requires WITH_BOOST, the library may not have been found. "
+			"Configure BOOST or disable WITH_INTERNATIONAL"
+		)
 	endif()
 endif()
 
 # See TEST_SSE_SUPPORT() for how this is defined.
 
-if(WITH_RAYOPTIMIZATION)
-	if(SUPPORT_SSE_BUILD)
-		set(PLATFORM_CFLAGS " ${COMPILER_SSE_FLAG} ${PLATFORM_CFLAGS}")
-		add_definitions(-D__SSE__ -D__MMX__)
-	endif()
-	if(SUPPORT_SSE2_BUILD)
-		set(PLATFORM_CFLAGS " ${COMPILER_SSE2_FLAG} ${PLATFORM_CFLAGS}")
-		add_definitions(-D__SSE2__)
-		if(NOT SUPPORT_SSE_BUILD) # dont double up
-			add_definitions(-D__MMX__)
-		endif()
+# Do it globally, SSE2 is required for quite some time now.
+# Doing it now allows to use SSE/SSE2 in inline headers.
+if(SUPPORT_SSE_BUILD)
+	set(PLATFORM_CFLAGS " ${COMPILER_SSE_FLAG} ${PLATFORM_CFLAGS}")
+	add_definitions(-D__SSE__ -D__MMX__)
+endif()
+if(SUPPORT_SSE2_BUILD)
+	set(PLATFORM_CFLAGS " ${COMPILER_SSE2_FLAG} ${PLATFORM_CFLAGS}")
+	add_definitions(-D__SSE2__)
+	if(NOT SUPPORT_SSE_BUILD) # dont double up
+		add_definitions(-D__MMX__)
 	endif()
 endif()
 
@@ -2410,7 +2544,10 @@ if(WITH_GL_PROFILE_COMPAT OR WITH_GL_PROFILE_CORE)
 elseif(WITH_GL_PROFILE_ES20)
 	if(WITH_SYSTEM_GLES)
 		if(NOT OPENGLES_LIBRARY)
-			message(FATAL_ERROR "Unable to find OpenGL ES libraries.  Install them or disable WITH_SYSTEM_GLES.")
+			message(FATAL_ERROR
+				"Unable to find OpenGL ES libraries. "
+				"Install them or disable WITH_SYSTEM_GLES."
+			)
 		endif()
 
 		list(APPEND BLENDER_GL_LIBRARIES OPENGLES_LIBRARY)
@@ -2422,7 +2559,10 @@ elseif(WITH_GL_PROFILE_ES20)
 		list(APPEND BLENDER_GL_LIBRARIES "${OPENGLES_LIBRARY}")
 
 		if(NOT OPENGLES_LIBRARY)
-			message(FATAL_ERROR "To compile WITH_GL_EGL you need to set OPENGLES_LIBRARY to the file path of an OpenGL ES 2.0 library.")
+			message(FATAL_ERROR
+				"To compile WITH_GL_EGL you need to set OPENGLES_LIBRARY "
+				"to the file path of an OpenGL ES 2.0 library."
+			)
 		endif()
 
 	endif()
@@ -2434,7 +2574,10 @@ elseif(WITH_GL_PROFILE_ES20)
 		mark_as_advanced(OPENGLES_DLL)
 
 		if(NOT OPENGLES_DLL)
-			message(FATAL_ERROR "To compile WITH_GL_PROFILE_ES20 you need to set OPENGLES_DLL to the file path of an OpenGL ES 2.0 runtime dynamic link library (DLL).")
+			message(FATAL_ERROR
+				"To compile WITH_GL_PROFILE_ES20 you need to set OPENGLES_DLL to the file "
+				"path of an OpenGL ES 2.0 runtime dynamic link library (DLL)."
+			)
 		endif()
 
 		if(WITH_GL_ANGLE)
@@ -2448,7 +2591,10 @@ elseif(WITH_GL_PROFILE_ES20)
 			mark_as_advanced(D3DCOMPILER_DLL)
 
 			if(D3DCOMPILER_DLL STREQUAL "")
-				message(FATAL_ERROR "To compile WITH_GL_ANGLE you need to set D3DCOMPILER_DLL to the file path of a copy of the DirectX redistributable DLL file: D3DCompiler_46.dll")
+				message(FATAL_ERROR
+					"To compile WITH_GL_ANGLE you need to set D3DCOMPILER_DLL to the file "
+					"path of a copy of the DirectX redistributable DLL file: D3DCompiler_46.dll"
+				)
 			endif()
 
 		endif()
@@ -2462,7 +2608,10 @@ if(WITH_GL_EGL)
 
 	if(WITH_SYSTEM_GLES)
 		if(NOT OPENGLES_EGL_LIBRARY)
-			message(FATAL_ERROR "Unable to find OpenGL ES libraries.  Install them or disable WITH_SYSTEM_GLES.")
+			message(FATAL_ERROR
+				"Unable to find OpenGL ES libraries. "
+				"Install them or disable WITH_SYSTEM_GLES."
+			)
 		endif()
 
 		list(APPEND BLENDER_GL_LIBRARIES OPENGLES_EGL_LIBRARY)
@@ -2474,7 +2623,10 @@ if(WITH_GL_EGL)
 		list(APPEND BLENDER_GL_LIBRARIES "${OPENGLES_LIBRARY}" "${OPENGLES_EGL_LIBRARY}")
 
 		if(NOT OPENGLES_EGL_LIBRARY)
-			message(FATAL_ERROR "To compile WITH_GL_EGL you need to set OPENGLES_EGL_LIBRARY to the file path of an EGL library.")
+			message(FATAL_ERROR
+				"To compile WITH_GL_EGL you need to set OPENGLES_EGL_LIBRARY "
+				"to the file path of an EGL library."
+			)
 		endif()
 
 	endif()
@@ -2486,7 +2638,10 @@ if(WITH_GL_EGL)
 		mark_as_advanced(OPENGLES_EGL_DLL)
 
 		if(NOT OPENGLES_EGL_DLL)
-			message(FATAL_ERROR "To compile WITH_GL_EGL you need to set OPENGLES_EGL_DLL to the file path of an EGL runtime dynamic link library (DLL).")
+			message(FATAL_ERROR
+				"To compile WITH_GL_EGL you need to set OPENGLES_EGL_DLL "
+				"to the file path of an EGL runtime dynamic link library (DLL)."
+			)
 		endif()
 
 	endif()
@@ -2651,7 +2806,9 @@ endif()
 if(WITH_LIBMV)
 	set(CERES_DEFINES)
 
-	if(SHARED_PTR_FOUND)
+	if(WITH_CXX11)
+		# nothing to be done
+	elseif(SHARED_PTR_FOUND)
 		if(SHARED_PTR_TR1_MEMORY_HEADER)
 			list(APPEND CERES_DEFINES -DCERES_TR1_MEMORY_HEADER)
 		endif()
@@ -2662,7 +2819,9 @@ if(WITH_LIBMV)
 		message(FATAL_ERROR "Ceres: Unable to find shared_ptr.")
 	endif()
 
-	if(HAVE_STD_UNORDERED_MAP_HEADER)
+	if(WITH_CXX11)
+		list(APPEND CERES_DEFINES -DCERES_STD_UNORDERED_MAP)
+	elseif(HAVE_STD_UNORDERED_MAP_HEADER)
 		if(HAVE_UNORDERED_MAP_IN_STD_NAMESPACE)
 			list(APPEND CERES_DEFINES -DCERES_STD_UNORDERED_MAP)
 		else()
@@ -2864,10 +3023,12 @@ endif()
 # be most problematic.
 if(WITH_PYTHON)
 	if(NOT EXISTS "${PYTHON_INCLUDE_DIR}/Python.h")
-		message(FATAL_ERROR "Missing: \"${PYTHON_INCLUDE_DIR}/Python.h\",\n"
-							"Set the cache entry 'PYTHON_INCLUDE_DIR' to point "
-							"to a valid python include path. Containing "
-							"Python.h for python version \"${PYTHON_VERSION}\"")
+		message(FATAL_ERROR
+			"Missing: \"${PYTHON_INCLUDE_DIR}/Python.h\",\n"
+			"Set the cache entry 'PYTHON_INCLUDE_DIR' to point "
+			"to a valid python include path. Containing "
+			"Python.h for python version \"${PYTHON_VERSION}\""
+		)
 	endif()
 
 	if(WIN32 OR APPLE)
@@ -3005,7 +3166,7 @@ if(FIRST_RUN)
 		string(LENGTH "${_msg}" _len)
 		while("32" GREATER "${_len}")
 			set(_msg "${_msg} ")
-			 math(EXPR _len "${_len} + 1")
+			math(EXPR _len "${_len} + 1")
 		endwhile()
 
 		set(_config_msg "${_config_msg}\n${_msg}${${_setting}}" PARENT_SCOPE)

GNUmakefile

@@ -87,13 +87,13 @@ ifndef NPROCS
 		NPROCS:=$(shell nproc)
 	endif
 	ifeq ($(OS), Darwin)
-		NPROCS:=$(shell sysctl -a | grep "hw.ncpu" | cut -d" " -f3)
+		NPROCS:=$(shell sysctl -n hw.ncpu)
 	endif
 	ifeq ($(OS), FreeBSD)
-		NPROCS:=$(shell sysctl -a | grep "hw.ncpu" | cut -d" " -f2 )
+		NPROCS:=$(shell sysctl -n hw.ncpu)
 	endif
 	ifeq ($(OS), NetBSD)
-		NPROCS:=$(shell sysctl -a | grep "hw.ncpu" | cut -d" " -f2 )
+		NPROCS:=$(shell sysctl -n hw.ncpu)
 	endif
 endif
 

build_files/build_environment/install_deps_patches/llvm.patch

@@ -1,12 +1,11 @@
 --- a/CMakeLists.txt
 +++ b/CMakeLists.txt
-@@ -13,7 +13,7 @@
- set(LLVM_VERSION_MAJOR 3)
- set(LLVM_VERSION_MINOR 1)
+@@ -14,7 +14,7 @@
+ set(LLVM_VERSION_MINOR 4)
  
--set(PACKAGE_VERSION "\${LLVM_VERSION_MAJOR}.\${LLVM_VERSION_MINOR}svn")
-+set(PACKAGE_VERSION "\${LLVM_VERSION_MAJOR}.\${LLVM_VERSION_MINOR}")
+ if (NOT PACKAGE_VERSION)
+-  set(PACKAGE_VERSION "${LLVM_VERSION_MAJOR}.${LLVM_VERSION_MINOR}svn")
++  set(PACKAGE_VERSION "${LLVM_VERSION_MAJOR}.${LLVM_VERSION_MINOR}")
+ endif()
  
- set_property(GLOBAL PROPERTY USE_FOLDERS ON)
- 
-
+ option(LLVM_INSTALL_TOOLCHAIN_ONLY "Only include toolchain files in the 'install' target." OFF)

build_files/buildbot/master.cfg

@@ -87,7 +87,7 @@ c['change_source'] = GitPoller(
 
 # CODEBASES
 #
-# Allow to controll separately things like branches for each repo and submodules.
+# Allow to control separately things like branches for each repo and submodules.
 
 all_repositories = {
     r'git://git.blender.org/blender.git': 'blender',

build_files/buildbot/slave_compile.py

@@ -56,7 +56,6 @@ if 'cmake' in builder:
     chroot_name = None  # If not None command will be delegated to that chroot
     cuda_chroot_name = None  # If not None cuda compilationcommand will be delegated to that chroot
     build_cubins = True  # Whether to build Cycles CUDA kernels
-    remove_install_dir = False  # Remove installation folder before building
     bits = 64
 
     # Config file to be used (relative to blender's sources root)
@@ -70,13 +69,12 @@ if 'cmake' in builder:
     cuda_cmake_options = []
 
     if builder.startswith('mac'):
-        install_dir = None
         # Set up OSX architecture
         if builder.endswith('x86_64_10_6_cmake'):
             cmake_extra_options.append('-DCMAKE_OSX_ARCHITECTURES:STRING=x86_64')
+        cmake_extra_options.append('-DCUDA_NVCC_EXECUTABLE=/usr/local/cuda-hack/bin/nvcc')
 
     elif builder.startswith('win'):
-        install_dir = None
         if builder.startswith('win64'):
             cmake_options.append(['-G', '"Visual Studio 12 2013 Win64"'])
         elif builder.startswith('win32'):
@@ -90,7 +88,6 @@ if 'cmake' in builder:
             deb_name = "jessie"
         elif glibc == 'glibc211':
             deb_name = "squeeze"
-        remove_install_dir = True
         cmake_config_file = "build_files/buildbot/config/blender_linux.cmake"
         cmake_player_config_file = "build_files/buildbot/config/blender_linux_player.cmake"
         if builder.endswith('x86_64_cmake'):
@@ -116,8 +113,7 @@ if 'cmake' in builder:
     if 'cuda' not in targets:
         cmake_options += cuda_cmake_options
 
-    if install_dir:
-        cmake_options.append("-DCMAKE_INSTALL_PREFIX=%s" % (install_dir))
+    cmake_options.append("-DCMAKE_INSTALL_PREFIX=%s" % (install_dir))
 
     cmake_options += cmake_extra_options
 
@@ -132,10 +128,8 @@ if 'cmake' in builder:
         cuda_chroot_prefix = chroot_prefix[:]
 
     # Make sure no garbage remained from the previous run
-    # (only do it if builder requested this)
-    if remove_install_dir:
-        if os.path.isdir(install_dir):
-            shutil.rmtree(install_dir)
+    if os.path.isdir(install_dir):
+        shutil.rmtree(install_dir)
 
     for target in targets:
         print("Building target %s" % (target))
@@ -187,7 +181,7 @@ if 'cmake' in builder:
             sys.exit(retcode)
 
         if builder.startswith('linux') and target == 'cuda':
-            blender_h = os.path.join(blender_dir, "source", "blender", "blenkernel", "BKE_blender.h")
+            blender_h = os.path.join(blender_dir, "source", "blender", "blenkernel", "BKE_blender_version.h")
             blender_version = int(parse_header_file(blender_h, 'BLENDER_VERSION'))
             blender_version = "%d.%d" % (blender_version // 100, blender_version % 100)
             kernels = os.path.join(target_build_dir, 'intern', 'cycles', 'kernel')

build_files/buildbot/slave_pack.py

@@ -130,7 +130,7 @@ if builder.find('cmake') != -1:
         blenderplayer = os.path.join(install_dir, 'blenderplayer')
 
         buildinfo_h = os.path.join(build_dir, "source", "creator", "buildinfo.h")
-        blender_h = os.path.join(blender_dir, "source", "blender", "blenkernel", "BKE_blender.h")
+        blender_h = os.path.join(blender_dir, "source", "blender", "blenkernel", "BKE_blender_version.h")
 
         # Get version information
         blender_version = int(parse_header_file(blender_h, 'BLENDER_VERSION'))

build_files/buildbot/slave_test.py

@@ -35,9 +35,8 @@ blender_dir = '../blender.git'
 if "cmake" in builder:
     # cmake
 
-    if "linux" in builder or 'mac' in builder:
-        print("Automated tests are still DISABLED!")
-        sys.exit(0)
+    print("Automated tests are still DISABLED!")
+    sys.exit(0)
 
     build_dir = os.path.abspath(os.path.join('..', 'build', builder))
     chroot_name = None

build_files/cmake/buildinfo.cmake

@@ -1,5 +1,10 @@
-# This is called by cmake as an extermal process from
+# This is called by cmake as an external process from
 # ./source/creator/CMakeLists.txt to write ./source/creator/buildinfo.h
+# Caller must define:
+#   SOURCE_DIR
+# Optional overrides:
+#   BUILD_DATE
+#   BUILD_TIME
 
 # Extract working copy information for SOURCE_DIR into MY_XXX variables
 # with a default in case anything fails, for example when using git-svn
@@ -128,12 +133,19 @@ endif()
 # BUILD_PLATFORM and BUILD_PLATFORM are taken from CMake
 # but BUILD_DATE and BUILD_TIME are platform dependent
 if(UNIX)
-	execute_process(COMMAND date "+%Y-%m-%d" OUTPUT_VARIABLE BUILD_DATE OUTPUT_STRIP_TRAILING_WHITESPACE)
-	execute_process(COMMAND date "+%H:%M:%S" OUTPUT_VARIABLE BUILD_TIME OUTPUT_STRIP_TRAILING_WHITESPACE)
-endif()
-if(WIN32)
-	execute_process(COMMAND cmd /c date /t OUTPUT_VARIABLE BUILD_DATE OUTPUT_STRIP_TRAILING_WHITESPACE)
-	execute_process(COMMAND cmd /c time /t OUTPUT_VARIABLE BUILD_TIME OUTPUT_STRIP_TRAILING_WHITESPACE)
+	if(NOT BUILD_DATE)
+		execute_process(COMMAND date "+%Y-%m-%d" OUTPUT_VARIABLE BUILD_DATE OUTPUT_STRIP_TRAILING_WHITESPACE)
+	endif()
+	if(NOT BUILD_TIME)
+		execute_process(COMMAND date "+%H:%M:%S" OUTPUT_VARIABLE BUILD_TIME OUTPUT_STRIP_TRAILING_WHITESPACE)
+	endif()
+elseif(WIN32)
+	if(NOT BUILD_DATE)
+		execute_process(COMMAND cmd /c date /t OUTPUT_VARIABLE BUILD_DATE OUTPUT_STRIP_TRAILING_WHITESPACE)
+	endif()
+	if(NOT BUILD_TIME)
+		execute_process(COMMAND cmd /c time /t OUTPUT_VARIABLE BUILD_TIME OUTPUT_STRIP_TRAILING_WHITESPACE)
+	endif()
 endif()
 
 # Write a file with the BUILD_HASH define

build_files/cmake/cmake_consistency_check.py

@@ -240,9 +240,10 @@ def cmake_get_src(f):
     filen.close()
 
 
-def is_ignore(f):
-    for ig in IGNORE:
+def is_ignore(f, ignore_used):
+    for index, ig in enumerate(IGNORE):
         if ig in f:
+            ignore_used[index] = True
             return True
     return False
 
@@ -283,10 +284,12 @@ def main():
     del is_err
     del errs
 
+    ignore_used = [False] * len(IGNORE)
+
     # now check on files not accounted for.
     print("\nC/C++ Files CMake doesnt know about...")
     for cf in sorted(source_list(SOURCE_DIR, is_c)):
-        if not is_ignore(cf):
+        if not is_ignore(cf, ignore_used):
             if cf not in global_c:
                 print("missing_c: ", cf)
 
@@ -303,7 +306,7 @@ def main():
 
     print("\nC/C++ Headers CMake doesnt know about...")
     for hf in sorted(source_list(SOURCE_DIR, is_c_header)):
-        if not is_ignore(hf):
+        if not is_ignore(hf, ignore_used):
             if hf not in global_h:
                 print("missing_h: ", hf)
 
@@ -324,5 +327,12 @@ def main():
                             if i > 1:
                                 traceback.print_exc()
 
+    # Check ignores aren't stale
+    print("\nCheck for unused 'IGNORE' paths...")
+    for index, ig in enumerate(IGNORE):
+        if not ignore_used[index]:
+            print("unused ignore: %r" % ig)
+
+
 if __name__ == "__main__":
     main()

build_files/cmake/cmake_consistency_check_config.py

@@ -3,11 +3,7 @@ import os
 IGNORE = (
     "/test/",
     "/tests/gtests/",
-    "/BSP_GhostTest/",
     "/release/",
-    "/xembed/",
-    "/TerraplayNetwork/",
-    "/ik_glut_test/",
 
     # specific source files
     "extern/bullet2/src/BulletCollision/CollisionDispatch/btBox2dBox2dCollisionAlgorithm.cpp",
@@ -18,31 +14,8 @@ IGNORE = (
     "extern/bullet2/src/BulletDynamics/Character/btKinematicCharacterController.cpp",
     "extern/bullet2/src/BulletDynamics/ConstraintSolver/btHinge2Constraint.cpp",
     "extern/bullet2/src/BulletDynamics/ConstraintSolver/btUniversalConstraint.cpp",
-    "extern/eltopo/common/meshes/ObjLoader.cpp",
-    "extern/eltopo/common/meshes/meshloader.cpp",
-    "extern/eltopo/common/openglutils.cpp",
-    "extern/eltopo/eltopo3d/broadphase_blenderbvh.cpp",
-    "source/blender/imbuf/intern/imbuf_cocoa.m",
-    "extern/recastnavigation/Recast/Source/RecastLog.cpp",
-    "extern/recastnavigation/Recast/Source/RecastTimer.cpp",
     "intern/audaspace/SRC/AUD_SRCResampleFactory.cpp",
     "intern/audaspace/SRC/AUD_SRCResampleReader.cpp",
-    "intern/cycles/render/film_response.cpp",
-    "extern/libmv/third_party/ceres/internal/ceres/generated/schur_eliminator_2_2_2.cc",
-    "extern/libmv/third_party/ceres/internal/ceres/generated/schur_eliminator_2_2_3.cc",
-    "extern/libmv/third_party/ceres/internal/ceres/generated/schur_eliminator_2_2_4.cc",
-    "extern/libmv/third_party/ceres/internal/ceres/generated/schur_eliminator_2_2_d.cc",
-    "extern/libmv/third_party/ceres/internal/ceres/generated/schur_eliminator_2_3_3.cc",
-    "extern/libmv/third_party/ceres/internal/ceres/generated/schur_eliminator_2_3_4.cc",
-    "extern/libmv/third_party/ceres/internal/ceres/generated/schur_eliminator_2_3_9.cc",
-    "extern/libmv/third_party/ceres/internal/ceres/generated/schur_eliminator_2_3_d.cc",
-    "extern/libmv/third_party/ceres/internal/ceres/generated/schur_eliminator_2_4_3.cc",
-    "extern/libmv/third_party/ceres/internal/ceres/generated/schur_eliminator_2_4_4.cc",
-    "extern/libmv/third_party/ceres/internal/ceres/generated/schur_eliminator_2_4_d.cc",
-    "extern/libmv/third_party/ceres/internal/ceres/generated/schur_eliminator_4_4_2.cc",
-    "extern/libmv/third_party/ceres/internal/ceres/generated/schur_eliminator_4_4_3.cc",
-    "extern/libmv/third_party/ceres/internal/ceres/generated/schur_eliminator_4_4_4.cc",
-    "extern/libmv/third_party/ceres/internal/ceres/generated/schur_eliminator_4_4_d.cc",
 
     "extern/bullet2/src/BulletCollision/CollisionDispatch/btBox2dBox2dCollisionAlgorithm.h",
     "extern/bullet2/src/BulletCollision/CollisionDispatch/btConvex2dConvex2dAlgorithm.h",
@@ -52,20 +25,12 @@ IGNORE = (
     "extern/bullet2/src/BulletDynamics/Character/btKinematicCharacterController.h",
     "extern/bullet2/src/BulletDynamics/ConstraintSolver/btHinge2Constraint.h",
     "extern/bullet2/src/BulletDynamics/ConstraintSolver/btUniversalConstraint.h",
-    "extern/eltopo/common/meshes/Edge.hpp",
-    "extern/eltopo/common/meshes/ObjLoader.hpp",
-    "extern/eltopo/common/meshes/TriangleIndex.hpp",
-    "extern/eltopo/common/meshes/meshloader.h",
-    "extern/eltopo/eltopo3d/broadphase_blenderbvh.h",
-    "extern/recastnavigation/Recast/Include/RecastLog.h",
-    "extern/recastnavigation/Recast/Include/RecastTimer.h",
+    "extern/carve/include/carve/config.h",
+    "extern/carve/include/carve/random/random.h",
+    "extern/carve/patches/files/config.h",
+    "extern/carve/patches/files/random.h",
     "intern/audaspace/SRC/AUD_SRCResampleFactory.h",
     "intern/audaspace/SRC/AUD_SRCResampleReader.h",
-    "intern/cycles/render/film_response.h",
-    "extern/carve/include/carve/config.h",
-    "extern/carve/include/carve/external/boost/random.hpp",
-    "extern/carve/patches/files/config.h",
-    "extern/carve/patches/files/random.hpp",
     )
 
 UTF8_CHECK = True

build_files/cmake/config/blender_full.cmake

@@ -54,7 +54,7 @@ set(WITH_PLAYER              ON  CACHE BOOL "" FORCE)
 set(WITH_MEM_JEMALLOC        ON  CACHE BOOL "" FORCE)
 
 
-# platform dependant options
+# platform dependent options
 if(UNIX AND NOT APPLE)
 	set(WITH_JACK                ON  CACHE BOOL "" FORCE)
 	set(WITH_DOC_MANPAGE         ON  CACHE BOOL "" FORCE)
@@ -69,7 +69,7 @@ elseif(WIN32)
 		set(WITH_CODEC_SNDFILE       OFF CACHE BOOL "" FORCE)
 		set(WITH_CYCLES_OSL          OFF CACHE BOOL "" FORCE)
 	endif()
-elseif (APPLE)
+elseif(APPLE)
 	set(WITH_JACK                ON  CACHE BOOL "" FORCE)
 	set(WITH_CODEC_QUICKTIME     ON  CACHE BOOL "" FORCE)
 	set(WITH_OPENSUBDIV          OFF CACHE BOOL "" FORCE)

build_files/cmake/config/blender_lite.cmake

@@ -14,6 +14,7 @@ set(WITH_CODEC_AVI           OFF CACHE BOOL "" FORCE)
 set(WITH_CODEC_FFMPEG        OFF CACHE BOOL "" FORCE)
 set(WITH_CODEC_SNDFILE       OFF CACHE BOOL "" FORCE)
 set(WITH_CYCLES              OFF CACHE BOOL "" FORCE)
+set(WITH_CYCLES_OSL          OFF CACHE BOOL "" FORCE)
 set(WITH_FFTW3               OFF CACHE BOOL "" FORCE)
 set(WITH_LIBMV               OFF CACHE BOOL "" FORCE)
 set(WITH_LLVM                OFF CACHE BOOL "" FORCE)

build_files/cmake/macros.cmake

@@ -141,22 +141,6 @@ function(target_link_libraries_debug
 	endforeach()
 endfunction()
 
-function(target_link_libraries_decoupled
-	target
-	libraries_var
-	)
-
-	if(NOT MSVC)
-		target_link_libraries(${target} ${${libraries_var}})
-	else()
-		# For MSVC we link to different libraries depending whether
-		# release or debug target is being built.
-		file_list_suffix(_libraries_debug "${${libraries_var}}" "_d")
-		target_link_libraries_debug(${target} "${_libraries_debug}")
-		target_link_libraries_optimized(${target} "${${libraries_var}}")
-	endif()
-endfunction()
-
 # Nicer makefiles with -I/1/foo/ instead of -I/1/2/3/../../foo/
 # use it instead of include_directories()
 function(blender_include_dirs
@@ -412,14 +396,7 @@ function(setup_liblinks
 	endif()
 	target_link_libraries(${target} ${JPEG_LIBRARIES})
 	if(WITH_IMAGE_OPENEXR)
-		if(WIN32 AND NOT UNIX AND NOT CMAKE_COMPILER_IS_GNUCC)
-			file_list_suffix(OPENEXR_LIBRARIES_DEBUG "${OPENEXR_LIBRARIES}" "_d")
-			target_link_libraries_debug(${target} "${OPENEXR_LIBRARIES_DEBUG}")
-			target_link_libraries_optimized(${target} "${OPENEXR_LIBRARIES}")
-			unset(OPENEXR_LIBRARIES_DEBUG)
-		else()
-			target_link_libraries(${target} ${OPENEXR_LIBRARIES})
-		endif()
+		target_link_libraries(${target} ${OPENEXR_LIBRARIES})
 	endif()
 	if(WITH_IMAGE_OPENJPEG AND WITH_SYSTEM_OPENJPEG)
 		target_link_libraries(${target} ${OPENJPEG_LIBRARIES})
@@ -601,6 +578,7 @@ function(SETUP_BLENDER_SORTED_LIBS)
 		ge_phys_bullet
 		bf_intern_smoke
 		extern_lzma
+		extern_curve_fit_nd
 		ge_logic_ketsji
 		extern_recastnavigation
 		ge_logic
@@ -990,6 +968,7 @@ macro(remove_strict_flags)
 		remove_cc_flag(
 			"-Wstrict-prototypes"
 			"-Wmissing-prototypes"
+			"-Wmissing-declarations"
 			"-Wmissing-format-attribute"
 			"-Wunused-local-typedefs"
 			"-Wunused-macros"
@@ -1071,6 +1050,19 @@ macro(remove_strict_flags_file
 
 endmacro()
 
+# External libs may need 'signed char' to be default.
+macro(remove_cc_flag_unsigned_char)
+	if(CMAKE_C_COMPILER_ID MATCHES "^(GNU|Clang|Intel)$")
+		remove_cc_flag("-funsigned-char")
+	elseif(MSVC)
+		remove_cc_flag("/J")
+	else()
+		message(WARNING
+			"Compiler '${CMAKE_C_COMPILER_ID}' failed to disable 'unsigned char' flag."
+			"Build files need updating."
+		)
+	endif()
+endmacro()
 
 function(ADD_CHECK_C_COMPILER_FLAG
 	_CFLAGS
@@ -1117,10 +1109,10 @@ function(get_blender_version)
 	# - BLENDER_VERSION_CYCLE (alpha, beta, rc, release)
 
 	# So cmake depends on BKE_blender.h, beware of inf-loops!
-	CONFIGURE_FILE(${CMAKE_SOURCE_DIR}/source/blender/blenkernel/BKE_blender.h
-	               ${CMAKE_BINARY_DIR}/source/blender/blenkernel/BKE_blender.h.done)
+	CONFIGURE_FILE(${CMAKE_SOURCE_DIR}/source/blender/blenkernel/BKE_blender_version.h
+	               ${CMAKE_BINARY_DIR}/source/blender/blenkernel/BKE_blender_version.h.done)
 
-	file(STRINGS ${CMAKE_SOURCE_DIR}/source/blender/blenkernel/BKE_blender.h _contents REGEX "^#define[ \t]+BLENDER_.*$")
+	file(STRINGS ${CMAKE_SOURCE_DIR}/source/blender/blenkernel/BKE_blender_version.h _contents REGEX "^#define[ \t]+BLENDER_.*$")
 
 	string(REGEX REPLACE ".*#define[ \t]+BLENDER_VERSION[ \t]+([0-9]+).*" "\\1" _out_version "${_contents}")
 	string(REGEX REPLACE ".*#define[ \t]+BLENDER_SUBVERSION[ \t]+([0-9]+).*" "\\1" _out_subversion "${_contents}")

doc/doxygen/doxygen.extern.h

@@ -11,6 +11,10 @@
  *
  */
 
+/** \defgroup curve_fit Curve Fitting Library
+ *  \ingroup extern
+ */
+
 /** \defgroup bullet Bullet Physics Library
  *  \ingroup extern
  *  \see \ref bulletdoc

doc/python_api/examples/bpy.props.3.py

@@ -20,7 +20,7 @@ bpy.types.Scene.my_settings = \
     bpy.props.CollectionProperty(type=SceneSettingItem)
 
 # Assume an armature object selected
-print("Adding 3 values!")
+print("Adding 2 values!")
 
 my_item = bpy.context.scene.my_settings.add()
 my_item.name = "Spam"

doc/python_api/examples/bpy.props.5.py

@@ -48,9 +48,9 @@ bpy.types.Scene.test_array = bpy.props.BoolVectorProperty(size=2, get=get_array,
 # Note: the getter/setter callback must use integer identifiers!
 test_items = [
     ("RED", "Red", "", 1),
-    ("GREEN", "Red", "", 2),
-    ("BLUE", "Red", "", 3),
-    ("YELLOW", "Red", "", 4),
+    ("GREEN", "Green", "", 2),
+    ("BLUE", "Blue", "", 3),
+    ("YELLOW", "Yellow", "", 4),
     ]
 
 

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

@@ -60,37 +60,37 @@ base class --- :class:`KX_GameObject`
 
       :type: float (read only)
 
-   ..attribute:: shadowFrustumSize
+   .. attribute:: shadowFrustumSize
 
       Size of the frustum used for creating the shadowmap.
 
       :type: float (read only)
 
-   ..attribute:: shadowBindId
+   .. attribute:: shadowBindId
 
       The OpenGL shadow texture bind number/id.
 
       :type: int (read only)
 
-   ..attribute:: shadowMapType
+   .. attribute:: shadowMapType
 
       The shadow shadow map type (0 -> Simple; 1 -> Variance)
 
       :type: int (read only)
 
-   ..attribute:: shadowBias
+   .. attribute:: shadowBias
 
       The shadow buffer sampling bias.
 
       :type: float (read only)
 
-   ..attribute:: shadowBleedBias
+   .. attribute:: shadowBleedBias
 
       The bias for reducing light-bleed on variance shadow maps.
 
       :type: float (read only)
 
-   ..attribute:: useShadow
+   .. attribute:: useShadow
 
       Returns True if the light has Shadow option activated, else returns False.
 

doc/python_api/sphinx_doc_gen.sh

@@ -33,10 +33,10 @@ SSH_UPLOAD="/data/www/vhosts/www.blender.org/api" # blender_python_api_VERSION,
 # "_".join(str(v) for v in bpy.app.version)
 # custom blender vars
 blender_srcdir=$(dirname -- $0)/../..
-blender_version=$(grep "BLENDER_VERSION\s" "$blender_srcdir/source/blender/blenkernel/BKE_blender.h" | awk '{print $3}')
-blender_version_char=$(grep "BLENDER_VERSION_CHAR\s" "$blender_srcdir/source/blender/blenkernel/BKE_blender.h" | awk '{print $3}')
-blender_version_cycle=$(grep "BLENDER_VERSION_CYCLE\s" "$blender_srcdir/source/blender/blenkernel/BKE_blender.h" | awk '{print $3}')
-blender_subversion=$(grep "BLENDER_SUBVERSION\s" "$blender_srcdir/source/blender/blenkernel/BKE_blender.h" | awk '{print $3}')
+blender_version=$(grep "BLENDER_VERSION\s" "$blender_srcdir/source/blender/blenkernel/BKE_blender_version.h" | awk '{print $3}')
+blender_version_char=$(grep "BLENDER_VERSION_CHAR\s" "$blender_srcdir/source/blender/blenkernel/BKE_blender_version.h" | awk '{print $3}')
+blender_version_cycle=$(grep "BLENDER_VERSION_CYCLE\s" "$blender_srcdir/source/blender/blenkernel/BKE_blender_version.h" | awk '{print $3}')
+blender_subversion=$(grep "BLENDER_SUBVERSION\s" "$blender_srcdir/source/blender/blenkernel/BKE_blender_version.h" | awk '{print $3}')
 
 if [ "$blender_version_cycle" = "release" ] ; then
 	BLENDER_VERSION=$(expr $blender_version / 100)_$(expr $blender_version % 100)$blender_version_char"_release"

extern/CMakeLists.txt

@@ -23,6 +23,9 @@
 #
 # ***** END GPL LICENSE BLOCK *****
 
+# Libs that adhere to strict flags
+add_subdirectory(curve_fit_nd)
+
 # Otherwise we get warnings here that we cant fix in external projects
 remove_strict_flags()
 

extern/bullet2/patches/blender.patch

@@ -233,3 +233,92 @@ index e05bdcc..dbcf2b6 100644
  
  
  #include "btVector3.h"
+diff --git a/extern/bullet2/src/BulletDynamics/Dynamics/btRigidBody.cpp b/extern/bullet2/src/BulletDynamics/Dynamics/btRigidBody.cpp
+index e0e8bc7..a788268 100644
+--- a/extern/bullet2/src/BulletDynamics/Dynamics/btRigidBody.cpp
++++ b/extern/bullet2/src/BulletDynamics/Dynamics/btRigidBody.cpp
+@@ -425,50 +425,38 @@ void btRigidBody::setCenterOfMassTransform(const btTransform& xform)
+ }
+ 
+ 
++bool btRigidBody::checkCollideWithOverride(const  btCollisionObject* co) const
++{
++	const btRigidBody* otherRb = btRigidBody::upcast(co);
++	if (!otherRb)
++		return true;
++
++	for (int i = 0; i < m_constraintRefs.size(); ++i)
++	{
++		const btTypedConstraint* c = m_constraintRefs[i];
++		if (c->isEnabled())
++			if (&c->getRigidBodyA() == otherRb || &c->getRigidBodyB() == otherRb)
++				return false;
++	}
++
++	return true;
++}
+ 
+ 
+ 
+ void btRigidBody::addConstraintRef(btTypedConstraint* c)
+ {
+-	///disable collision with the 'other' body
+-
+ 	int index = m_constraintRefs.findLinearSearch(c);
+-	//don't add constraints that are already referenced
+-	//btAssert(index == m_constraintRefs.size());
+ 	if (index == m_constraintRefs.size())
+-	{
+-		m_constraintRefs.push_back(c);
+-		btCollisionObject* colObjA = &c->getRigidBodyA();
+-		btCollisionObject* colObjB = &c->getRigidBodyB();
+-		if (colObjA == this)
+-		{
+-			colObjA->setIgnoreCollisionCheck(colObjB, true);
+-		}
+-		else
+-		{
+-			colObjB->setIgnoreCollisionCheck(colObjA, true);
+-		}
+-	} 
++		m_constraintRefs.push_back(c); 
++
++	m_checkCollideWith = true;
+ }
+ 
+ void btRigidBody::removeConstraintRef(btTypedConstraint* c)
+ {
+-	int index = m_constraintRefs.findLinearSearch(c);
+-	//don't remove constraints that are not referenced
+-	if(index < m_constraintRefs.size())
+-    {
+-        m_constraintRefs.remove(c);
+-        btCollisionObject* colObjA = &c->getRigidBodyA();
+-        btCollisionObject* colObjB = &c->getRigidBodyB();
+-        if (colObjA == this)
+-        {
+-            colObjA->setIgnoreCollisionCheck(colObjB, false);
+-        }
+-        else
+-        {
+-            colObjB->setIgnoreCollisionCheck(colObjA, false);
+-        }
+-    }
++	m_constraintRefs.remove(c);
++	m_checkCollideWith = m_constraintRefs.size() > 0;
+ }
+ 
+ int	btRigidBody::calculateSerializeBufferSize()	const
+diff --git a/extern/bullet2/src/BulletDynamics/Dynamics/btRigidBody.h b/extern/bullet2/src/BulletDynamics/Dynamics/btRigidBody.h
+index 1d177db..c2f8c5d 100644
+--- a/extern/bullet2/src/BulletDynamics/Dynamics/btRigidBody.h
++++ b/extern/bullet2/src/BulletDynamics/Dynamics/btRigidBody.h
+@@ -509,6 +509,8 @@ public:
+ 		return (getBroadphaseProxy() != 0);
+ 	}
+ 
++	virtual bool checkCollideWithOverride(const  btCollisionObject* co) const;
++
+ 	void addConstraintRef(btTypedConstraint* c);
+ 	void removeConstraintRef(btTypedConstraint* c);
+ 

extern/bullet2/src/BulletDynamics/Dynamics/btRigidBody.cpp

@@ -425,50 +425,38 @@ void btRigidBody::setCenterOfMassTransform(const btTransform& xform)
 }
 
 
+bool btRigidBody::checkCollideWithOverride(const  btCollisionObject* co) const
+{
+	const btRigidBody* otherRb = btRigidBody::upcast(co);
+	if (!otherRb)
+		return true;
+
+	for (int i = 0; i < m_constraintRefs.size(); ++i)
+	{
+		const btTypedConstraint* c = m_constraintRefs[i];
+		if (c->isEnabled())
+			if (&c->getRigidBodyA() == otherRb || &c->getRigidBodyB() == otherRb)
+				return false;
+	}
+
+	return true;
+}
 
 
 
 void btRigidBody::addConstraintRef(btTypedConstraint* c)
 {
-	///disable collision with the 'other' body
-
 	int index = m_constraintRefs.findLinearSearch(c);
-	//don't add constraints that are already referenced
-	//btAssert(index == m_constraintRefs.size());
 	if (index == m_constraintRefs.size())
-	{
-		m_constraintRefs.push_back(c);
-		btCollisionObject* colObjA = &c->getRigidBodyA();
-		btCollisionObject* colObjB = &c->getRigidBodyB();
-		if (colObjA == this)
-		{
-			colObjA->setIgnoreCollisionCheck(colObjB, true);
-		}
-		else
-		{
-			colObjB->setIgnoreCollisionCheck(colObjA, true);
-		}
-	} 
+		m_constraintRefs.push_back(c); 
+
+	m_checkCollideWith = true;
 }
 
 void btRigidBody::removeConstraintRef(btTypedConstraint* c)
 {
-	int index = m_constraintRefs.findLinearSearch(c);
-	//don't remove constraints that are not referenced
-	if(index < m_constraintRefs.size())
-    {
-        m_constraintRefs.remove(c);
-        btCollisionObject* colObjA = &c->getRigidBodyA();
-        btCollisionObject* colObjB = &c->getRigidBodyB();
-        if (colObjA == this)
-        {
-            colObjA->setIgnoreCollisionCheck(colObjB, false);
-        }
-        else
-        {
-            colObjB->setIgnoreCollisionCheck(colObjA, false);
-        }
-    }
+	m_constraintRefs.remove(c);
+	m_checkCollideWith = m_constraintRefs.size() > 0;
 }
 
 int	btRigidBody::calculateSerializeBufferSize()	const

extern/bullet2/src/BulletDynamics/Dynamics/btRigidBody.h

@@ -509,6 +509,8 @@ public:
 		return (getBroadphaseProxy() != 0);
 	}
 
+	virtual bool checkCollideWithOverride(const  btCollisionObject* co) const;
+
 	void addConstraintRef(btTypedConstraint* c);
 	void removeConstraintRef(btTypedConstraint* c);
 

extern/bullet2/src/BulletDynamics/Featherstone/btMultiBody.cpp

@@ -51,6 +51,7 @@ namespace {
         bottom_out = -displacement.cross(top_out) + rotation_matrix * bottom_in;
     }
 
+/*
     void InverseSpatialTransform(const btMatrix3x3 &rotation_matrix,
                                  const btVector3 &displacement,
                                  const btVector3 &top_in,
@@ -80,6 +81,7 @@ namespace {
 		top_out = a_top.cross(b_top);
 		bottom_out = a_bottom.cross(b_top) + a_top.cross(b_bottom);
 	}
+*/
 }
 
 

extern/bullet2/src/BulletSoftBody/btSoftBodyHelpers.cpp

@@ -1127,6 +1127,7 @@ int								nattrb=0;
 int								hasbounds=0;
 int result = sscanf(node,"%d %d %d %d",&nnode,&ndims,&nattrb,&hasbounds);
 result = sscanf(node,"%d %d %d %d",&nnode,&ndims,&nattrb,&hasbounds);
+(void)result;
 node += nextLine(node);
 
 pos.resize(nnode);
@@ -1208,10 +1209,10 @@ if(ele&&ele[0])
 			}
 		}
 	}
-printf("Nodes:  %u\r\n",psb->m_nodes.size());
-printf("Links:  %u\r\n",psb->m_links.size());
-printf("Faces:  %u\r\n",psb->m_faces.size());
-printf("Tetras: %u\r\n",psb->m_tetras.size());
+printf("Nodes:  %d\r\n",psb->m_nodes.size());
+printf("Links:  %d\r\n",psb->m_links.size());
+printf("Faces:  %d\r\n",psb->m_faces.size());
+printf("Tetras: %d\r\n",psb->m_tetras.size());
 return(psb);
 }
 

extern/bullet2/src/BulletSoftBody/btSoftBodyInternals.h

@@ -422,7 +422,7 @@ static inline btVector3		BaryCoord(	const btVector3& a,
 }
 
 //
-static btScalar				ImplicitSolve(	btSoftBody::ImplicitFn* fn,
+static inline btScalar				ImplicitSolve(	btSoftBody::ImplicitFn* fn,
 										  const btVector3& a,
 										  const btVector3& b,
 										  const btScalar accuracy,
@@ -504,7 +504,7 @@ static inline btScalar		VolumeOf(	const btVector3& x0,
 }
 
 //
-static void					EvaluateMedium(	const btSoftBodyWorldInfo* wfi,
+static inline void					EvaluateMedium(	const btSoftBodyWorldInfo* wfi,
 										   const btVector3& x,
 										   btSoftBody::sMedium& medium)
 {

extern/carve/lib/geom2d.cpp

@@ -157,6 +157,7 @@ namespace carve {
       return pointInPoly(points, p2_adapt_ident(), p);
     }
 
+#if 0
     static int lineSegmentPolyIntersections(const P2Vector &points,
                                             LineSegment2 line,
                                             std::vector<PolyIntersectionInfo> &out) {
@@ -224,6 +225,7 @@ namespace carve {
       }
       return count;
     }
+#endif
  
     struct FwdSort {
       bool operator()(const PolyIntersectionInfo &a,
@@ -239,6 +241,7 @@ namespace carve {
       }
     };
 
+#if 0
     static int sortedLineSegmentPolyIntersections(const P2Vector &points,
                                                   LineSegment2 line,
                                                   std::vector<PolyIntersectionInfo> &out) {
@@ -253,6 +256,7 @@ namespace carve {
       }
       return count;
     }
+#endif
 
     bool pickContainedPoint(const std::vector<P2> &poly, P2 &result) {
       return pickContainedPoint(poly, p2_adapt_ident(), result);

extern/carve/lib/geom3d.cpp

@@ -28,6 +28,7 @@ namespace carve {
   namespace geom3d {
 
     namespace {
+#if 0
       int is_same(const std::vector<const Vector *> &a,
           const std::vector<const Vector *> &b) {
         if (a.size() != b.size()) return false;
@@ -52,6 +53,7 @@ not_fwd:
 not_rev:
         return 0;
       }
+#endif
     }
 
     bool planeIntersection(const Plane &a, const Plane &b, Ray &r) {

extern/carve/lib/intersect.cpp

@@ -236,6 +236,7 @@ namespace {
 
 
 
+#if 0
   void dump_intersections(std::ostream &out, carve::csg::Intersections &csg_intersections) {
     std::vector<dump_data> temp;
 
@@ -284,13 +285,14 @@ namespace {
         vertices.push_back(i_pt->v);
       }
     }
+#endif
 
     carve::point::PointSet points(vertices);
 
     std::string outf("/tmp/intersection-points.ply");
     ::writePLY(outf, &points, true);
-#endif
   }
+#endif
 
 
 
@@ -481,6 +483,7 @@ void carve::csg::CSG::makeVertexIntersections() {
 
 
 
+#if 0
 static carve::mesh::MeshSet<3>::vertex_t *chooseWeldPoint(
     const carve::csg::detail::VSet &equivalent,
     carve::csg::VertexPool &vertex_pool) {
@@ -537,7 +540,7 @@ static const carve::mesh::MeshSet<3>::vertex_t *weld(
   }
   return weld_point;
 }
-
+#endif
 
 
 void carve::csg::CSG::groupIntersections() {
@@ -1219,6 +1222,7 @@ void carve::csg::CSG::makeFaceEdges(carve::csg::EdgeClassification &eclass,
  * 
  * @param fll 
  */
+#if 0
 static void checkFaceLoopIntegrity(carve::csg::FaceLoopList &fll) {
   static carve::TimingName FUNC_NAME("CSG::checkFaceLoopIntegrity()");
   carve::TimingBlock block(FUNC_NAME);
@@ -1245,7 +1249,7 @@ static void checkFaceLoopIntegrity(carve::csg::FaceLoopList &fll) {
     }
   }
 }
-
+#endif
 
 
 /** 

extern/carve/lib/intersect_classify_edge.cpp

@@ -139,6 +139,7 @@ namespace carve {
 
 
 
+#if 0
       static void walkGraphSegment(carve::csg::detail::VVSMap &shared_edge_graph,
                                    const carve::csg::detail::VSet &branch_points,
                                    V2 initial,
@@ -215,7 +216,7 @@ namespace carve {
 #endif
 #endif
       }
-
+#endif
 
 
       static carve::geom3d::Vector perpendicular(const carve::geom3d::Vector &v) {
@@ -383,6 +384,7 @@ namespace carve {
 
 
 
+#if 0
       static void traceIntersectionGraph(const V2Set &shared_edges,
                                          const FLGroupList & /* a_loops_grouped */,
                                          const FLGroupList & /* b_loops_grouped */,
@@ -416,6 +418,7 @@ namespace carve {
           walkGraphSegment(shared_edge_graph, branch_points, V2(v1, v2), a_edge_map, b_edge_map, out);
         }
       }
+#endif
 
       void hashByPerimeter(FLGroupList &grp, PerimMap &perim_map) {
         for (FLGroupList::iterator i = grp.begin(); i != grp.end(); ++i) {

extern/carve/lib/intersect_face_division.cpp

@@ -1409,6 +1409,7 @@ namespace {
     return s.str().substr(1);
   }
 
+#if 0
   void dumpAsGraph(carve::mesh::MeshSet<3>::face_t *face,
                    const std::vector<carve::mesh::MeshSet<3>::vertex_t *> &base_loop,
                    const carve::csg::V2Set &face_edges,
@@ -1450,6 +1451,7 @@ namespace {
     }
     std::cerr << "};\n";
   }
+#endif
 
   void generateOneFaceLoop(carve::mesh::MeshSet<3>::face_t *face,
                            const carve::csg::detail::Data &data,

extern/carve/lib/math.cpp

@@ -43,6 +43,7 @@ namespace carve {
     };
 
     namespace {
+#if 0
       void cplx_sqrt(double re, double im,
                      double &re_1, double &im_1,
                      double &re_2, double &im_2) {
@@ -57,7 +58,9 @@ namespace carve {
           im_2 = -im_1;
         }
       }
+#endif
 
+#if 0
     void cplx_cbrt(double re, double im,
                    double &re_1, double &im_1,
                    double &re_2, double &im_2,
@@ -76,6 +79,7 @@ namespace carve {
         im_3 = r * sin(t + M_TWOPI * 2.0 / 3.0);
       }
     }
+#endif
 
       void add_root(std::vector<Root> &roots, double root) {
         for (size_t i = 0; i < roots.size(); ++i) {
@@ -250,6 +254,7 @@ namespace carve {
       e2.normalize();
     }
 
+#if 0
     static void eig3(const Matrix3 &m,
                      double l,
                      carve::geom::vector<3> &e1,
@@ -259,6 +264,7 @@ namespace carve {
       e2.x = 0.0; e2.y = 1.0; e2.z = 0.0;
       e3.x = 0.0; e3.y = 0.0; e3.z = 1.0;
     }
+#endif
 
     void eigSolveSymmetric(const Matrix3 &m,
                            double &l1, carve::geom::vector<3> &e1,

extern/carve/lib/mesh.cpp

@@ -774,7 +774,6 @@ namespace carve {
   // connectivity information in the Polyhedron.
   mesh::MeshSet<3> *meshFromPolyhedron(const poly::Polyhedron *poly, int manifold_id) {
     typedef mesh::Vertex<3> vertex_t;
-    typedef mesh::Vertex<3>::vector_t vector_t;
     typedef mesh::Edge<3> edge_t;
     typedef mesh::Face<3> face_t;
     typedef mesh::Mesh<3> mesh_t;
@@ -884,7 +883,6 @@ namespace carve {
 
   // construct a Polyhedron from a MeshSet
   poly::Polyhedron *polyhedronFromMesh(const mesh::MeshSet<3> *mesh, int manifold_id) {
-    typedef poly::Polyhedron poly_t;
     typedef poly::Polyhedron::vertex_t vertex_t;
     typedef poly::Polyhedron::edge_t edge_t;
     typedef poly::Polyhedron::face_t face_t;

extern/cuew/CMakeLists.txt

@@ -33,8 +33,9 @@ set(INC_SYS
 )
 
 set(SRC
-	include/cuew.h
 	src/cuew.c
+
+	include/cuew.h
 )
 
 blender_add_lib(extern_cuew "${SRC}" "${INC}" "${INC_SYS}")

extern/cuew/auto/cuda_errors.py

@@ -1,35 +0,0 @@
-CUDA_ERRORS={
-'CUDA_SUCCESS': "No errors",
-'CUDA_ERROR_INVALID_VALUE': "Invalid value",
-'CUDA_ERROR_OUT_OF_MEMORY': "Out of memory",
-'CUDA_ERROR_NOT_INITIALIZED': "Driver not initialized",
-'CUDA_ERROR_DEINITIALIZED': "Driver deinitialized",
-'CUDA_ERROR_NO_DEVICE': "No CUDA-capable device available",
-'CUDA_ERROR_INVALID_DEVICE': "Invalid device",
-'CUDA_ERROR_INVALID_IMAGE': "Invalid kernel image",
-'CUDA_ERROR_INVALID_CONTEXT': "Invalid context",
-'CUDA_ERROR_CONTEXT_ALREADY_CURRENT': "Context already current",
-'CUDA_ERROR_MAP_FAILED': "Map failed",
-'CUDA_ERROR_UNMAP_FAILED': "Unmap failed",
-'CUDA_ERROR_ARRAY_IS_MAPPED': "Array is mapped",
-'CUDA_ERROR_ALREADY_MAPPED': "Already mapped",
-'CUDA_ERROR_NO_BINARY_FOR_GPU': "No binary for GPU",
-'CUDA_ERROR_ALREADY_ACQUIRED': "Already acquired",
-'CUDA_ERROR_NOT_MAPPED': "Not mapped",
-'CUDA_ERROR_NOT_MAPPED_AS_ARRAY': "Mapped resource not available for access as an array",
-'CUDA_ERROR_NOT_MAPPED_AS_POINTER': "Mapped resource not available for access as a pointer",
-'CUDA_ERROR_ECC_UNCORRECTABLE': "Uncorrectable ECC error detected",
-'CUDA_ERROR_UNSUPPORTED_LIMIT': "CUlimit not supported by device",
-'CUDA_ERROR_INVALID_SOURCE': "Invalid source",
-'CUDA_ERROR_FILE_NOT_FOUND': "File not found",
-'CUDA_ERROR_SHARED_OBJECT_SYMBOL_NOT_FOUND': "Link to a shared object failed to resolve",
-'CUDA_ERROR_SHARED_OBJECT_INIT_FAILED': "Shared object initialization failed",
-'CUDA_ERROR_INVALID_HANDLE': "Invalid handle",
-'CUDA_ERROR_NOT_FOUND': "Not found",
-'CUDA_ERROR_NOT_READY': "CUDA not ready",
-'CUDA_ERROR_LAUNCH_FAILED': "Launch failed",
-'CUDA_ERROR_LAUNCH_OUT_OF_RESOURCES': "Launch exceeded resources",
-'CUDA_ERROR_LAUNCH_TIMEOUT': "Launch exceeded timeout",
-'CUDA_ERROR_LAUNCH_INCOMPATIBLE_TEXTURING': "Launch with incompatible texturing",
-'CUDA_ERROR_UNKNOWN': "Unknown error",
-}

extern/cuew/auto/cuda_extra.py

@@ -1,125 +0,0 @@
-extra_code = """
-static void path_join(const char *path1,
-                      const char *path2,
-                      int maxlen,
-                      char *result) {
-#if defined(WIN32) || defined(_WIN32)
-  const char separator = '\\\\';
-#else
-  const char separator = '/';
-#endif
-  int n = snprintf(result, maxlen, "%s%c%s", path1, separator, path2);
-  if (n != -1 && n < maxlen) {
-    result[n] = '\\0';
-  }
-  else {
-    result[maxlen - 1] = '\\0';
-  }
-}
-
-static int path_exists(const char *path) {
-  struct stat st;
-  if (stat(path, &st)) {
-    return 0;
-  }
-  return 1;
-}
-
-const char *cuewCompilerPath(void) {
-#ifdef _WIN32
-  const char *defaultpaths[] = {"C:/CUDA/bin", NULL};
-  const char *executable = "nvcc.exe";
-#else
-  const char *defaultpaths[] = {
-    "/Developer/NVIDIA/CUDA-5.0/bin",
-    "/usr/local/cuda-5.0/bin",
-    "/usr/local/cuda/bin",
-    "/Developer/NVIDIA/CUDA-6.0/bin",
-    "/usr/local/cuda-6.0/bin",
-    "/Developer/NVIDIA/CUDA-5.5/bin",
-    "/usr/local/cuda-5.5/bin",
-    NULL};
-  const char *executable = "nvcc";
-#endif
-  int i;
-
-  const char *binpath = getenv("CUDA_BIN_PATH");
-
-  static char nvcc[65536];
-
-  if (binpath) {
-    path_join(binpath, executable, sizeof(nvcc), nvcc);
-    if (path_exists(nvcc))
-      return nvcc;
-  }
-
-  for (i = 0; defaultpaths[i]; ++i) {
-    path_join(defaultpaths[i], executable, sizeof(nvcc), nvcc);
-    if (path_exists(nvcc))
-      return nvcc;
-  }
-
-#ifndef _WIN32
-  {
-    FILE *handle = popen("which nvcc", "r");
-    if (handle) {
-      char buffer[4096] = {0};
-      int len = fread(buffer, 1, sizeof(buffer) - 1, handle);
-      buffer[len] = '\\0';
-      pclose(handle);
-
-      if (buffer[0])
-        return "nvcc";
-    }
-  }
-#endif
-
-  return NULL;
-}
-
-int cuewCompilerVersion(void) {
-  const char *path = cuewCompilerPath();
-  const char *marker = "Cuda compilation tools, release ";
-  FILE *pipe;
-  int major, minor;
-  char *versionstr;
-  char buf[128];
-  char output[65536] = "\\0";
-  char command[65536] = "\\0";
-
-  if (path == NULL)
-    return 0;
-
-  /* get --version output */
-  strncpy(command, path, sizeof(command));
-  strncat(command, " --version", sizeof(command) - strlen(path));
-  pipe = popen(command, "r");
-  if (!pipe) {
-    fprintf(stderr, "CUDA: failed to run compiler to retrieve version");
-    return 0;
-  }
-
-  while (!feof(pipe)) {
-    if (fgets(buf, sizeof(buf), pipe) != NULL) {
-      strncat(output, buf, sizeof(output) - strlen(output) - 1);
-    }
-  }
-
-  pclose(pipe);
-
-  /* parse version number */
-  versionstr = strstr(output, marker);
-  if (versionstr == NULL) {
-    fprintf(stderr, "CUDA: failed to find version number in:\\n\\n%s\\n", output);
-    return 0;
-  }
-  versionstr += strlen(marker);
-
-  if (sscanf(versionstr, "%d.%d", &major, &minor) < 2) {
-    fprintf(stderr, "CUDA: failed to parse version number from:\\n\\n%s\\n", output);
-    return 0;
-  }
-
-  return 10 * major + minor;
-}
-"""

extern/cuew/auto/cuew_gen.py

@@ -1,640 +0,0 @@
-#!/usr/bin/env python3
-#
-# Copyright 2014 Blender Foundation
-#
-# Licensed under the Apache License, Version 2.0 (the "License");
-# you may not use this file except in compliance with the License.
-# You may obtain a copy of the License at
-#
-# http://www.apache.org/licenses/LICENSE-2.0
-#
-# Unless required by applicable law or agreed to in writing, software
-# distributed under the License is distributed on an "AS IS" BASIS,
-# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
-# See the License for the specific language governing permissions and
-# limitations under the License
-
-# This script generates either header or implementation file from
-# a CUDA header files.
-#
-# Usage: cuew hdr|impl [/path/to/cuda/includes]
-#  - hdr means header file will be generated and printed to stdout.
-#  - impl means implementation file will be generated and printed to stdout.
-#  - /path/to/cuda/includes is a path to a folder with cuda.h and cudaGL.h
-#    for which wrangler will be generated.
-
-import os
-import sys
-from cuda_errors import CUDA_ERRORS
-from pycparser import c_parser, c_ast, parse_file
-from subprocess import Popen, PIPE
-
-INCLUDE_DIR = "/usr/include"
-LIB = "CUEW"
-REAL_LIB = "CUDA"
-VERSION_MAJOR = "1"
-VERSION_MINOR = "2"
-COPYRIGHT = """/*
- * Copyright 2011-2014 Blender Foundation
- *
- * Licensed under the Apache License, Version 2.0 (the "License");
- * you may not use this file except in compliance with the License.
- * You may obtain a copy of the License at
- *
- * http://www.apache.org/licenses/LICENSE-2.0
- *
- * Unless required by applicable law or agreed to in writing, software
- * distributed under the License is distributed on an "AS IS" BASIS,
- * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
- * See the License for the specific language governing permissions and
- * limitations under the License
- */"""
-FILES = ["cuda.h", "cudaGL.h", 'nvrtc.h']
-
-TYPEDEFS = []
-FUNC_TYPEDEFS = []
-SYMBOLS = []
-DEFINES = []
-DEFINES_V2 = []
-ERRORS = []
-
-
-class FuncDefVisitor(c_ast.NodeVisitor):
-    indent = 0
-    prev_complex = False
-    dummy_typedefs = ['size_t', 'CUdeviceptr']
-
-    def _get_quals_string(self, node):
-        if node.quals:
-            return ' '.join(node.quals) + ' '
-        return ''
-
-    def _get_ident_type(self, node):
-        if isinstance(node, c_ast.PtrDecl):
-            return self._get_ident_type(node.type.type) + '*'
-        if isinstance(node, c_ast.ArrayDecl):
-            return self._get_ident_type(node.type)
-        elif isinstance(node, c_ast.Struct):
-            if node.name:
-                return 'struct ' + node.name
-            else:
-                self.indent += 1
-                struct = self._stringify_struct(node)
-                self.indent -= 1
-                return "struct {\n" + \
-                       struct + ("  " * self.indent) + "}"
-        elif isinstance(node, c_ast.Union):
-            self.indent += 1
-            union = self._stringify_struct(node)
-            self.indent -= 1
-            return "union {\n" + union + ("  " * self.indent) + "}"
-        elif isinstance(node, c_ast.Enum):
-            if node.name is not None:
-                return 'enum ' + node.name
-            else:
-                return 'enum '
-        elif isinstance(node, c_ast.TypeDecl):
-            return self._get_ident_type(node.type)
-        else:
-            return node.names[0]
-
-    def _stringify_param(self, param):
-        param_type = param.type
-        result = self._get_quals_string(param)
-        result += self._get_ident_type(param_type)
-        if param.name:
-            result += ' ' + param.name
-        if isinstance(param_type, c_ast.ArrayDecl):
-            # TODO(sergey): Workaround to deal with the
-            # preprocessed file where array size got
-            # substituded.
-            dim = param_type.dim.value
-            if param.name == "reserved" and dim == "64":
-                dim = "CU_IPC_HANDLE_SIZE"
-            result += '[' + dim + ']'
-        return result
-
-    def _stringify_params(self, params):
-        result = []
-        for param in params:
-            result.append(self._stringify_param(param))
-        return ', '.join(result)
-
-    def _stringify_struct(self, node):
-        result = ""
-        children = node.children()
-        for child in children:
-            member = self._stringify_param(child[1])
-            result += ("  " * self.indent) + member + ";\n"
-        return result
-
-    def _stringify_enum(self, node):
-        result = ""
-        children = node.children()
-        for child in children:
-            if isinstance(child[1], c_ast.EnumeratorList):
-                enumerators = child[1].enumerators
-                for enumerator in enumerators:
-                    result += ("  " * self.indent) + enumerator.name
-                    if enumerator.value:
-                        result += " = " + enumerator.value.value
-                    result += ",\n"
-                    if enumerator.name.startswith("CUDA_ERROR_"):
-                        ERRORS.append(enumerator.name)
-        return result
-
-    def visit_Decl(self, node):
-        if node.type.__class__.__name__ == 'FuncDecl':
-            if isinstance(node.type, c_ast.FuncDecl):
-                func_decl = node.type
-                func_decl_type = func_decl.type
-
-                typedef = 'typedef '
-                symbol_name = None
-
-                if isinstance(func_decl_type, c_ast.TypeDecl):
-                    symbol_name = func_decl_type.declname
-                    typedef += self._get_quals_string(func_decl_type)
-                    typedef += self._get_ident_type(func_decl_type.type)
-                    typedef += ' CUDAAPI'
-                    typedef += ' t' + symbol_name
-                elif isinstance(func_decl_type, c_ast.PtrDecl):
-                    ptr_type = func_decl_type.type
-                    symbol_name = ptr_type.declname
-                    typedef += self._get_quals_string(ptr_type)
-                    typedef += self._get_ident_type(func_decl_type)
-                    typedef += ' CUDAAPI'
-                    typedef += ' t' + symbol_name
-
-                typedef += '(' + \
-                    self._stringify_params(func_decl.args.params) + \
-                    ');'
-
-                SYMBOLS.append(symbol_name)
-                FUNC_TYPEDEFS.append(typedef)
-
-    def visit_Typedef(self, node):
-        if node.name in self.dummy_typedefs:
-            return
-
-        complex = False
-        type = self._get_ident_type(node.type)
-        quals = self._get_quals_string(node)
-
-        if isinstance(node.type.type, c_ast.Struct):
-            self.indent += 1
-            struct = self._stringify_struct(node.type.type)
-            self.indent -= 1
-            typedef = quals + type + " {\n" + struct + "} " + node.name
-            complex = True
-        elif isinstance(node.type.type, c_ast.Enum):
-            self.indent += 1
-            enum = self._stringify_enum(node.type.type)
-            self.indent -= 1
-            typedef = quals + type + " {\n" + enum + "} " + node.name
-            complex = True
-        else:
-            typedef = quals + type + " " + node.name
-        if complex or self.prev_complex:
-            typedef = "\ntypedef " + typedef + ";"
-        else:
-            typedef = "typedef " + typedef + ";"
-
-        TYPEDEFS.append(typedef)
-
-        self.prev_complex = complex
-
-
-def get_latest_cpp():
-    path_prefix = "/usr/bin"
-    for cpp_version in ["9", "8", "7", "6", "5", "4"]:
-        test_cpp = os.path.join(path_prefix, "cpp-4." + cpp_version)
-        if os.path.exists(test_cpp):
-            return test_cpp
-    return None
-
-
-def preprocess_file(filename, cpp_path):
-    args = [cpp_path, "-I./"]
-    if filename.endswith("GL.h"):
-        args.append("-DCUDAAPI= ")
-    args.append(filename)
-
-    try:
-        pipe = Popen(args,
-                     stdout=PIPE,
-                     universal_newlines=True)
-        text = pipe.communicate()[0]
-    except OSError as e:
-        raise RuntimeError("Unable to invoke 'cpp'.  " +
-            'Make sure its path was passed correctly\n' +
-            ('Original error: %s' % e))
-
-    return text
-
-
-def parse_files():
-    parser = c_parser.CParser()
-    cpp_path = get_latest_cpp()
-
-    for filename in FILES:
-        filepath = os.path.join(INCLUDE_DIR, filename)
-        dummy_typedefs = {}
-        text = preprocess_file(filepath, cpp_path)
-
-        if filepath.endswith("GL.h"):
-            dummy_typedefs = {
-                "CUresult": "int",
-                "CUgraphicsResource": "void *",
-                "CUdevice": "void *",
-                "CUcontext": "void *",
-                "CUdeviceptr": "void *",
-                "CUstream": "void *"
-                }
-
-            text = "typedef int GLint;\n" + text
-            text = "typedef unsigned int GLuint;\n" + text
-            text = "typedef unsigned int GLenum;\n" + text
-            text = "typedef long size_t;\n" + text
-
-        for typedef in sorted(dummy_typedefs):
-            text = "typedef " + dummy_typedefs[typedef] + " " + \
-                typedef + ";\n" + text
-
-        ast = parser.parse(text, filepath)
-
-        with open(filepath) as f:
-            lines = f.readlines()
-            for line in lines:
-                if line.startswith("#define"):
-                    line = line[8:-1]
-                    token = line.split()
-                    if token[0] not in ("__cuda_cuda_h__",
-                                        "CUDA_CB",
-                                        "CUDAAPI",
-                                        "CUDAGL_H",
-                                        "__NVRTC_H__"):
-                        DEFINES.append(token)
-
-            for line in lines:
-                # TODO(sergey): Use better matching rule for _v2 symbols.
-                if line[0].isspace() and line.lstrip().startswith("#define"):
-                    line = line[12:-1]
-                    token = line.split()
-                    if len(token) == 2 and (token[1].endswith("_v2") or
-                                            token[1].endswith("_v2)")):
-                        if token[1].startswith('__CUDA_API_PTDS') or \
-                           token[1].startswith('__CUDA_API_PTSZ'):
-                            token[1] = token[1][16:-1]
-                        DEFINES_V2.append(token)
-
-        v = FuncDefVisitor()
-        for typedef in dummy_typedefs:
-            v.dummy_typedefs.append(typedef)
-        v.visit(ast)
-
-        FUNC_TYPEDEFS.append('')
-        SYMBOLS.append('')
-
-
-def print_copyright():
-    print(COPYRIGHT)
-    print("")
-
-
-def open_header_guard():
-    print("#ifndef __%s_H__" % (LIB))
-    print("#define __%s_H__" % (LIB))
-    print("")
-    print("#ifdef __cplusplus")
-    print("extern \"C\" {")
-    print("#endif")
-    print("")
-
-
-def close_header_guard():
-    print("")
-    print("#ifdef __cplusplus")
-    print("}")
-    print("#endif")
-    print("")
-    print("#endif  /* __%s_H__ */" % (LIB))
-
-
-def print_header():
-    print_copyright()
-    open_header_guard()
-
-    # Fot size_t.
-    print("#include <stdlib.h>")
-    print("")
-
-    print("/* Defines. */")
-    print("#define %s_VERSION_MAJOR %s" % (LIB, VERSION_MAJOR))
-    print("#define %s_VERSION_MINOR %s" % (LIB, VERSION_MINOR))
-    print("")
-    for define in DEFINES:
-        print('#define %s' % (' '.join(define)))
-    print("")
-
-    print("""/* Functions which changed 3.1 -> 3.2 for 64 bit stuff,
- * the cuda library has both the old ones for compatibility and new
- * ones with _v2 postfix,
- */""")
-    for define in DEFINES_V2:
-        print('#define %s' % (' '.join(define)))
-    print("")
-
-    print("/* Types. */")
-
-    # We handle this specially because of the file is
-    # getting preprocessed.
-    print("""#if defined(__x86_64) || defined(AMD64) || defined(_M_AMD64)
-typedef unsigned long long CUdeviceptr;
-#else
-typedef unsigned int CUdeviceptr;
-#endif
-""")
-
-    for typedef in TYPEDEFS:
-        print('%s' % (typedef))
-
-    # TDO(sergey): This is only specific to CUDA wrapper.
-    print("""
-#ifdef _WIN32
-#  define CUDAAPI __stdcall
-#  define CUDA_CB __stdcall
-#else
-#  define CUDAAPI
-#  define CUDA_CB
-#endif
-""")
-
-    print("/* Function types. */")
-    for func_typedef in FUNC_TYPEDEFS:
-        print('%s' % (func_typedef))
-    print("")
-
-    print("/* Function declarations. */")
-    for symbol in SYMBOLS:
-        if symbol:
-            print('extern t%s *%s;' % (symbol, symbol))
-        else:
-            print("")
-
-    print("")
-    print("enum {")
-    print("  CUEW_SUCCESS = 0,")
-    print("  CUEW_ERROR_OPEN_FAILED = -1,")
-    print("  CUEW_ERROR_ATEXIT_FAILED = -2,")
-    print("};")
-    print("")
-    print("int %sInit(void);" % (LIB.lower()))
-    # TODO(sergey): Get rid of hardcoded CUresult.
-    print("const char *%sErrorString(CUresult result);" % (LIB.lower()))
-    print("const char *cuewCompilerPath(void);")
-    print("int cuewCompilerVersion(void);")
-
-    close_header_guard()
-
-
-def print_dl_wrapper():
-    print("""#ifdef _WIN32
-#  define WIN32_LEAN_AND_MEAN
-#  define VC_EXTRALEAN
-#  include <windows.h>
-
-/* Utility macros. */
-
-typedef HMODULE DynamicLibrary;
-
-#  define dynamic_library_open(path)         LoadLibraryA(path)
-#  define dynamic_library_close(lib)         FreeLibrary(lib)
-#  define dynamic_library_find(lib, symbol)  GetProcAddress(lib, symbol)
-#else
-#  include <dlfcn.h>
-
-typedef void* DynamicLibrary;
-
-#  define dynamic_library_open(path)         dlopen(path, RTLD_NOW)
-#  define dynamic_library_close(lib)         dlclose(lib)
-#  define dynamic_library_find(lib, symbol)  dlsym(lib, symbol)
-#endif
-""")
-
-
-def print_dl_helper_macro():
-    print("""#define _LIBRARY_FIND_CHECKED(lib, name) \\
-        name = (t##name *)dynamic_library_find(lib, #name); \\
-        assert(name);
-
-#define _LIBRARY_FIND(lib, name) \\
-        name = (t##name *)dynamic_library_find(lib, #name);
-
-#define %s_LIBRARY_FIND_CHECKED(name) \\
-        _LIBRARY_FIND_CHECKED(cuda_lib, name)
-#define %s_LIBRARY_FIND(name) _LIBRARY_FIND(cuda_lib, name)
-
-#define NVRTC_LIBRARY_FIND_CHECKED(name) \\
-        _LIBRARY_FIND_CHECKED(nvrtc_lib, name)
-#define NVRTC_LIBRARY_FIND(name) _LIBRARY_FIND(nvrtc_lib, name)
-
-static DynamicLibrary cuda_lib;
-static DynamicLibrary nvrtc_lib;""" % (REAL_LIB, REAL_LIB))
-    print("")
-
-
-def print_dl_helpers():
-    print("""static DynamicLibrary dynamic_library_open_find(const char **paths) {
-  int i = 0;
-  while (paths[i] != NULL) {
-      DynamicLibrary lib = dynamic_library_open(paths[i]);
-      if (lib != NULL) {
-        return lib;
-      }
-      ++i;
-  }
-  return NULL;
-}
-
-static void %sExit(void) {
-  if(cuda_lib != NULL) {
-    /*  Ignore errors. */
-    dynamic_library_close(cuda_lib);
-    cuda_lib = NULL;
-  }
-}""" % (LIB.lower()))
-    print("")
-
-
-def print_lib_path():
-    # TODO(sergey): get rid of hardcoded libraries.
-    print("""#ifdef _WIN32
-  /* Expected in c:/windows/system or similar, no path needed. */
-  const char *cuda_paths[] = {"nvcuda.dll", NULL};
-  const char *nvrtc_paths[] = {"nvrtc.dll", NULL};
-#elif defined(__APPLE__)
-  /* Default installation path. */
-  const char *cuda_paths[] = {"/usr/local/cuda/lib/libcuda.dylib", NULL};
-  const char *nvrtc_paths[] = {"/usr/local/cuda/lib/libnvrtc.dylib", NULL};
-#else
-  const char *cuda_paths[] = {"libcuda.so", NULL};
-  const char *nvrtc_paths[] = {"libnvrtc.so",
-#  if defined(__x86_64__) || defined(_M_X64)
-                               "/usr/local/cuda/lib64/libnvrtc.so",
-#else
-                               "/usr/local/cuda/lib/libnvrtc.so",
-#endif
-                               NULL};
-#endif""")
-
-
-def print_init_guard():
-    print("""  static int initialized = 0;
-  static int result = 0;
-  int error, driver_version;
-
-  if (initialized) {
-    return result;
-  }
-
-  initialized = 1;
-
-  error = atexit(cuewExit);
-  if (error) {
-    result = CUEW_ERROR_ATEXIT_FAILED;
-    return result;
-  }
-
-  /* Load library. */
-  cuda_lib = dynamic_library_open_find(cuda_paths);
-  nvrtc_lib = dynamic_library_open_find(nvrtc_paths);
-
-  /* CUDA library is mandatory to have, while nvrtc might be missing. */
-  if (cuda_lib == NULL) {
-    result = CUEW_ERROR_OPEN_FAILED;
-    return result;
-  }""")
-    print("")
-
-
-def print_driver_version_guard():
-    # TODO(sergey): Currently it's hardcoded for CUDA only.
-    print("""  /* Detect driver version. */
-  driver_version = 1000;
-
-  %s_LIBRARY_FIND_CHECKED(cuDriverGetVersion);
-  if (cuDriverGetVersion) {
-    cuDriverGetVersion(&driver_version);
-  }
-
-  /* We require version 4.0. */
-  if (driver_version < 4000) {
-    result = CUEW_ERROR_OPEN_FAILED;
-    return result;
-  }""" % (REAL_LIB))
-
-
-def print_dl_init():
-    print("int %sInit(void) {" % (LIB.lower()))
-
-    print("  /* Library paths. */")
-    print_lib_path()
-    print_init_guard()
-    print_driver_version_guard()
-
-    print("  /* Fetch all function pointers. */")
-    for symbol in SYMBOLS:
-        if symbol:
-          if not symbol.startswith('nvrtc'):
-            print("  %s_LIBRARY_FIND(%s);" % (REAL_LIB, symbol))
-        else:
-            print("")
-
-    print("  if (nvrtc_lib != NULL) {")
-    for symbol in SYMBOLS:
-        if symbol and symbol.startswith('nvrtc'):
-            print("    NVRTC_LIBRARY_FIND(%s);" % (symbol))
-    print("  }")
-
-    print("")
-    print("  result = CUEW_SUCCESS;")
-    print("  return result;")
-
-    print("}")
-
-
-def print_implementation():
-    print_copyright()
-
-    # TODO(sergey): Get rid of hardcoded header.
-    print("""#ifdef _MSC_VER
-#  define snprintf _snprintf
-#  define popen _popen
-#  define pclose _pclose
-#  define _CRT_SECURE_NO_WARNINGS
-#endif
-""")
-    print("#include <cuew.h>")
-    print("#include <assert.h>")
-    print("#include <stdio.h>")
-    print("#include <string.h>")
-    print("#include <sys/stat.h>")
-    print("")
-
-    print_dl_wrapper()
-    print_dl_helper_macro()
-
-    print("/* Function definitions. */")
-    for symbol in SYMBOLS:
-        if symbol:
-            print('t%s *%s;' % (symbol, symbol))
-        else:
-            print("")
-    print("")
-
-    print_dl_helpers()
-
-    print("/* Implementation function. */")
-    print_dl_init()
-
-    print("")
-    # TODO(sergey): Get rid of hardcoded CUresult.
-    print("const char *%sErrorString(CUresult result) {" % (LIB.lower()))
-    print("  switch(result) {")
-    print("    case CUDA_SUCCESS: return \"No errors\";")
-
-    for error in ERRORS:
-        if error in CUDA_ERRORS:
-            str = CUDA_ERRORS[error]
-        else:
-            temp = error[11:].replace('_', ' ')
-            str = temp[0] + temp[1:].lower()
-        print("    case %s: return \"%s\";" % (error, str))
-
-    print("    default: return \"Unknown CUDA error value\";")
-    print("  }")
-    print("}")
-
-    from cuda_extra import extra_code
-    print(extra_code)
-
-if __name__ == "__main__":
-
-    if len(sys.argv) != 2 and len(sys.argv) != 3:
-        print("Usage: %s hdr|impl [/path/to/cuda/toolkit/include]" %
-              (sys.argv[0]))
-        exit(1)
-
-    if len(sys.argv) == 3:
-        INCLUDE_DIR = sys.argv[2]
-
-    parse_files()
-
-    if sys.argv[1] == "hdr":
-        print_header()
-    elif sys.argv[1] == "impl":
-        print_implementation()
-    else:
-        print("Unknown command %s" % (sys.argv[1]))
-        exit(1)

extern/cuew/auto/cuew_gen.sh

@@ -1,10 +0,0 @@
-#!/bin/sh
-
-# This script invokes cuew_gen.py and updates the
-# header and source files in the repository.
-
-SCRIPT=`realpath -s $0`
-DIR=`dirname $SCRIPT`
-
-python ${DIR}/cuew_gen.py hdr $@ > $DIR/../include/cuew.h
-python ${DIR}/cuew_gen.py impl $@ > $DIR/../src/cuew.c

extern/cuew/auto/stdlib.h

@@ -1,3 +0,0 @@
-/* This file is needed to workaround issue with parsing system headers. */
-
-typedef long size_t;

extern/cuew/include/cuew.h

@@ -131,8 +131,8 @@ typedef struct CUsurfref_st* CUsurfref;
 typedef struct CUevent_st* CUevent;
 typedef struct CUstream_st* CUstream;
 typedef struct CUgraphicsResource_st* CUgraphicsResource;
-typedef unsigned CUtexObject;
-typedef unsigned CUsurfObject;
+typedef unsigned long long CUtexObject;
+typedef unsigned long long CUsurfObject;
 
 typedef struct CUuuid_st {
   char bytes[16];
@@ -603,7 +603,7 @@ typedef struct CUDA_ARRAY_DESCRIPTOR_st {
   size_t Width;
   size_t Height;
   CUarray_format Format;
-  unsigned NumChannels;
+  unsigned int NumChannels;
 } CUDA_ARRAY_DESCRIPTOR;
 
 typedef struct CUDA_ARRAY3D_DESCRIPTOR_st {
@@ -611,8 +611,8 @@ typedef struct CUDA_ARRAY3D_DESCRIPTOR_st {
   size_t Height;
   size_t Depth;
   CUarray_format Format;
-  unsigned NumChannels;
-  unsigned Flags;
+  unsigned int NumChannels;
+  unsigned int Flags;
 } CUDA_ARRAY3D_DESCRIPTOR;
 
 typedef struct CUDA_RESOURCE_DESC_st {
@@ -627,13 +627,13 @@ typedef struct CUDA_RESOURCE_DESC_st {
     struct {
       CUdeviceptr devPtr;
       CUarray_format format;
-      unsigned numChannels;
+      unsigned int numChannels;
       size_t sizeInBytes;
     } linear;
     struct {
       CUdeviceptr devPtr;
       CUarray_format format;
-      unsigned numChannels;
+      unsigned int numChannels;
       size_t width;
       size_t height;
       size_t pitchInBytes;
@@ -642,14 +642,14 @@ typedef struct CUDA_RESOURCE_DESC_st {
       int reserved[32];
     } reserved;
   } res;
-  unsigned flags;
+  unsigned int flags;
 } CUDA_RESOURCE_DESC;
 
 typedef struct CUDA_TEXTURE_DESC_st {
   CUaddress_mode addressMode[3];
   CUfilter_mode filterMode;
-  unsigned flags;
-  unsigned maxAnisotropy;
+  unsigned int flags;
+  unsigned int maxAnisotropy;
   CUfilter_mode mipmapFilterMode;
   float mipmapLevelBias;
   float minMipmapLevelClamp;
@@ -700,19 +700,19 @@ typedef struct CUDA_RESOURCE_VIEW_DESC_st {
   size_t width;
   size_t height;
   size_t depth;
-  unsigned firstMipmapLevel;
-  unsigned lastMipmapLevel;
-  unsigned firstLayer;
-  unsigned lastLayer;
-  unsigned reserved[16];
+  unsigned int firstMipmapLevel;
+  unsigned int lastMipmapLevel;
+  unsigned int firstLayer;
+  unsigned int lastLayer;
+  unsigned int reserved[16];
 } CUDA_RESOURCE_VIEW_DESC;
 
 typedef struct CUDA_POINTER_ATTRIBUTE_P2P_TOKENS_st {
-  unsigned p2pToken;
-  unsigned vaSpaceToken;
+  unsigned long long p2pToken;
+  unsigned int vaSpaceToken;
 } CUDA_POINTER_ATTRIBUTE_P2P_TOKENS;
-typedef unsigned GLenum;
-typedef unsigned GLuint;
+typedef unsigned int GLenum;
+typedef unsigned int GLuint;
 typedef int GLint;
 
 typedef enum CUGLDeviceList_enum {
@@ -751,7 +751,7 @@ typedef struct _nvrtcProgram* nvrtcProgram;
 /* Function types. */
 typedef CUresult CUDAAPI tcuGetErrorString(CUresult error, const char* pStr);
 typedef CUresult CUDAAPI tcuGetErrorName(CUresult error, const char* pStr);
-typedef CUresult CUDAAPI tcuInit(unsigned Flags);
+typedef CUresult CUDAAPI tcuInit(unsigned int Flags);
 typedef CUresult CUDAAPI tcuDriverGetVersion(int* driverVersion);
 typedef CUresult CUDAAPI tcuDeviceGet(CUdevice* device, int ordinal);
 typedef CUresult CUDAAPI tcuDeviceGetCount(int* count);
@@ -762,17 +762,17 @@ typedef CUresult CUDAAPI tcuDeviceGetProperties(CUdevprop* prop, CUdevice dev);
 typedef CUresult CUDAAPI tcuDeviceComputeCapability(int* major, int* minor, CUdevice dev);
 typedef CUresult CUDAAPI tcuDevicePrimaryCtxRetain(CUcontext* pctx, CUdevice dev);
 typedef CUresult CUDAAPI tcuDevicePrimaryCtxRelease(CUdevice dev);
-typedef CUresult CUDAAPI tcuDevicePrimaryCtxSetFlags(CUdevice dev, unsigned flags);
-typedef CUresult CUDAAPI tcuDevicePrimaryCtxGetState(CUdevice dev, unsigned* flags, int* active);
+typedef CUresult CUDAAPI tcuDevicePrimaryCtxSetFlags(CUdevice dev, unsigned int flags);
+typedef CUresult CUDAAPI tcuDevicePrimaryCtxGetState(CUdevice dev, unsigned int* flags, int* active);
 typedef CUresult CUDAAPI tcuDevicePrimaryCtxReset(CUdevice dev);
-typedef CUresult CUDAAPI tcuCtxCreate_v2(CUcontext* pctx, unsigned flags, CUdevice dev);
+typedef CUresult CUDAAPI tcuCtxCreate_v2(CUcontext* pctx, unsigned int flags, CUdevice dev);
 typedef CUresult CUDAAPI tcuCtxDestroy_v2(CUcontext ctx);
 typedef CUresult CUDAAPI tcuCtxPushCurrent_v2(CUcontext ctx);
 typedef CUresult CUDAAPI tcuCtxPopCurrent_v2(CUcontext* pctx);
 typedef CUresult CUDAAPI tcuCtxSetCurrent(CUcontext ctx);
 typedef CUresult CUDAAPI tcuCtxGetCurrent(CUcontext* pctx);
 typedef CUresult CUDAAPI tcuCtxGetDevice(CUdevice* device);
-typedef CUresult CUDAAPI tcuCtxGetFlags(unsigned* flags);
+typedef CUresult CUDAAPI tcuCtxGetFlags(unsigned int* flags);
 typedef CUresult CUDAAPI tcuCtxSynchronize(void);
 typedef CUresult CUDAAPI tcuCtxSetLimit(CUlimit limit, size_t value);
 typedef CUresult CUDAAPI tcuCtxGetLimit(size_t* pvalue, CUlimit limit);
@@ -780,43 +780,43 @@ typedef CUresult CUDAAPI tcuCtxGetCacheConfig(CUfunc_cache* pconfig);
 typedef CUresult CUDAAPI tcuCtxSetCacheConfig(CUfunc_cache config);
 typedef CUresult CUDAAPI tcuCtxGetSharedMemConfig(CUsharedconfig* pConfig);
 typedef CUresult CUDAAPI tcuCtxSetSharedMemConfig(CUsharedconfig config);
-typedef CUresult CUDAAPI tcuCtxGetApiVersion(CUcontext ctx, unsigned* version);
+typedef CUresult CUDAAPI tcuCtxGetApiVersion(CUcontext ctx, unsigned int* version);
 typedef CUresult CUDAAPI tcuCtxGetStreamPriorityRange(int* leastPriority, int* greatestPriority);
-typedef CUresult CUDAAPI tcuCtxAttach(CUcontext* pctx, unsigned flags);
+typedef CUresult CUDAAPI tcuCtxAttach(CUcontext* pctx, unsigned int flags);
 typedef CUresult CUDAAPI tcuCtxDetach(CUcontext ctx);
 typedef CUresult CUDAAPI tcuModuleLoad(CUmodule* module, const char* fname);
 typedef CUresult CUDAAPI tcuModuleLoadData(CUmodule* module, const void* image);
-typedef CUresult CUDAAPI tcuModuleLoadDataEx(CUmodule* module, const void* image, unsigned numOptions, CUjit_option* options, void* optionValues);
+typedef CUresult CUDAAPI tcuModuleLoadDataEx(CUmodule* module, const void* image, unsigned int numOptions, CUjit_option* options, void* optionValues);
 typedef CUresult CUDAAPI tcuModuleLoadFatBinary(CUmodule* module, const void* fatCubin);
 typedef CUresult CUDAAPI tcuModuleUnload(CUmodule hmod);
 typedef CUresult CUDAAPI tcuModuleGetFunction(CUfunction* hfunc, CUmodule hmod, const char* name);
 typedef CUresult CUDAAPI tcuModuleGetGlobal_v2(CUdeviceptr* dptr, size_t* bytes, CUmodule hmod, const char* name);
 typedef CUresult CUDAAPI tcuModuleGetTexRef(CUtexref* pTexRef, CUmodule hmod, const char* name);
 typedef CUresult CUDAAPI tcuModuleGetSurfRef(CUsurfref* pSurfRef, CUmodule hmod, const char* name);
-typedef CUresult CUDAAPI tcuLinkCreate_v2(unsigned numOptions, CUjit_option* options, void* optionValues, CUlinkState* stateOut);
-typedef CUresult CUDAAPI tcuLinkAddData_v2(CUlinkState state, CUjitInputType type, void* data, size_t size, const char* name, unsigned numOptions, CUjit_option* options, void* optionValues);
-typedef CUresult CUDAAPI tcuLinkAddFile_v2(CUlinkState state, CUjitInputType type, const char* path, unsigned numOptions, CUjit_option* options, void* optionValues);
+typedef CUresult CUDAAPI tcuLinkCreate_v2(unsigned int numOptions, CUjit_option* options, void* optionValues, CUlinkState* stateOut);
+typedef CUresult CUDAAPI tcuLinkAddData_v2(CUlinkState state, CUjitInputType type, void* data, size_t size, const char* name, unsigned int numOptions, CUjit_option* options, void* optionValues);
+typedef CUresult CUDAAPI tcuLinkAddFile_v2(CUlinkState state, CUjitInputType type, const char* path, unsigned int numOptions, CUjit_option* options, void* optionValues);
 typedef CUresult CUDAAPI tcuLinkComplete(CUlinkState state, void* cubinOut, size_t* sizeOut);
 typedef CUresult CUDAAPI tcuLinkDestroy(CUlinkState state);
 typedef CUresult CUDAAPI tcuMemGetInfo_v2(size_t* free, size_t* total);
 typedef CUresult CUDAAPI tcuMemAlloc_v2(CUdeviceptr* dptr, size_t bytesize);
-typedef CUresult CUDAAPI tcuMemAllocPitch_v2(CUdeviceptr* dptr, size_t* pPitch, size_t WidthInBytes, size_t Height, unsigned ElementSizeBytes);
+typedef CUresult CUDAAPI tcuMemAllocPitch_v2(CUdeviceptr* dptr, size_t* pPitch, size_t WidthInBytes, size_t Height, unsigned int ElementSizeBytes);
 typedef CUresult CUDAAPI tcuMemFree_v2(CUdeviceptr dptr);
 typedef CUresult CUDAAPI tcuMemGetAddressRange_v2(CUdeviceptr* pbase, size_t* psize, CUdeviceptr dptr);
 typedef CUresult CUDAAPI tcuMemAllocHost_v2(void* pp, size_t bytesize);
 typedef CUresult CUDAAPI tcuMemFreeHost(void* p);
-typedef CUresult CUDAAPI tcuMemHostAlloc(void* pp, size_t bytesize, unsigned Flags);
-typedef CUresult CUDAAPI tcuMemHostGetDevicePointer_v2(CUdeviceptr* pdptr, void* p, unsigned Flags);
-typedef CUresult CUDAAPI tcuMemHostGetFlags(unsigned* pFlags, void* p);
-typedef CUresult CUDAAPI tcuMemAllocManaged(CUdeviceptr* dptr, size_t bytesize, unsigned flags);
+typedef CUresult CUDAAPI tcuMemHostAlloc(void* pp, size_t bytesize, unsigned int Flags);
+typedef CUresult CUDAAPI tcuMemHostGetDevicePointer_v2(CUdeviceptr* pdptr, void* p, unsigned int Flags);
+typedef CUresult CUDAAPI tcuMemHostGetFlags(unsigned int* pFlags, void* p);
+typedef CUresult CUDAAPI tcuMemAllocManaged(CUdeviceptr* dptr, size_t bytesize, unsigned int flags);
 typedef CUresult CUDAAPI tcuDeviceGetByPCIBusId(CUdevice* dev, const char* pciBusId);
 typedef CUresult CUDAAPI tcuDeviceGetPCIBusId(char* pciBusId, int len, CUdevice dev);
 typedef CUresult CUDAAPI tcuIpcGetEventHandle(CUipcEventHandle* pHandle, CUevent event);
 typedef CUresult CUDAAPI tcuIpcOpenEventHandle(CUevent* phEvent, CUipcEventHandle handle);
 typedef CUresult CUDAAPI tcuIpcGetMemHandle(CUipcMemHandle* pHandle, CUdeviceptr dptr);
-typedef CUresult CUDAAPI tcuIpcOpenMemHandle(CUdeviceptr* pdptr, CUipcMemHandle handle, unsigned Flags);
+typedef CUresult CUDAAPI tcuIpcOpenMemHandle(CUdeviceptr* pdptr, CUipcMemHandle handle, unsigned int Flags);
 typedef CUresult CUDAAPI tcuIpcCloseMemHandle(CUdeviceptr dptr);
-typedef CUresult CUDAAPI tcuMemHostRegister_v2(void* p, size_t bytesize, unsigned Flags);
+typedef CUresult CUDAAPI tcuMemHostRegister_v2(void* p, size_t bytesize, unsigned int Flags);
 typedef CUresult CUDAAPI tcuMemHostUnregister(void* p);
 typedef CUresult CUDAAPI tcuMemcpy(CUdeviceptr dst, CUdeviceptr src, size_t ByteCount);
 typedef CUresult CUDAAPI tcuMemcpyPeer(CUdeviceptr dstDevice, CUcontext dstContext, CUdeviceptr srcDevice, CUcontext srcContext, size_t ByteCount);
@@ -842,40 +842,40 @@ typedef CUresult CUDAAPI tcuMemcpyAtoHAsync_v2(void* dstHost, CUarray srcArray,
 typedef CUresult CUDAAPI tcuMemcpy2DAsync_v2(const CUDA_MEMCPY2D* pCopy, CUstream hStream);
 typedef CUresult CUDAAPI tcuMemcpy3DAsync_v2(const CUDA_MEMCPY3D* pCopy, CUstream hStream);
 typedef CUresult CUDAAPI tcuMemcpy3DPeerAsync(const CUDA_MEMCPY3D_PEER* pCopy, CUstream hStream);
-typedef CUresult CUDAAPI tcuMemsetD8_v2(CUdeviceptr dstDevice, unsigned uc, size_t N);
-typedef CUresult CUDAAPI tcuMemsetD16_v2(CUdeviceptr dstDevice, unsigned us, size_t N);
-typedef CUresult CUDAAPI tcuMemsetD32_v2(CUdeviceptr dstDevice, unsigned ui, size_t N);
-typedef CUresult CUDAAPI tcuMemsetD2D8_v2(CUdeviceptr dstDevice, size_t dstPitch, unsigned uc, size_t Width, size_t Height);
-typedef CUresult CUDAAPI tcuMemsetD2D16_v2(CUdeviceptr dstDevice, size_t dstPitch, unsigned us, size_t Width, size_t Height);
-typedef CUresult CUDAAPI tcuMemsetD2D32_v2(CUdeviceptr dstDevice, size_t dstPitch, unsigned ui, size_t Width, size_t Height);
-typedef CUresult CUDAAPI tcuMemsetD8Async(CUdeviceptr dstDevice, unsigned uc, size_t N, CUstream hStream);
-typedef CUresult CUDAAPI tcuMemsetD16Async(CUdeviceptr dstDevice, unsigned us, size_t N, CUstream hStream);
-typedef CUresult CUDAAPI tcuMemsetD32Async(CUdeviceptr dstDevice, unsigned ui, size_t N, CUstream hStream);
-typedef CUresult CUDAAPI tcuMemsetD2D8Async(CUdeviceptr dstDevice, size_t dstPitch, unsigned uc, size_t Width, size_t Height, CUstream hStream);
-typedef CUresult CUDAAPI tcuMemsetD2D16Async(CUdeviceptr dstDevice, size_t dstPitch, unsigned us, size_t Width, size_t Height, CUstream hStream);
-typedef CUresult CUDAAPI tcuMemsetD2D32Async(CUdeviceptr dstDevice, size_t dstPitch, unsigned ui, size_t Width, size_t Height, CUstream hStream);
+typedef CUresult CUDAAPI tcuMemsetD8_v2(CUdeviceptr dstDevice, unsigned char uc, size_t N);
+typedef CUresult CUDAAPI tcuMemsetD16_v2(CUdeviceptr dstDevice, unsigned short us, size_t N);
+typedef CUresult CUDAAPI tcuMemsetD32_v2(CUdeviceptr dstDevice, unsigned int ui, size_t N);
+typedef CUresult CUDAAPI tcuMemsetD2D8_v2(CUdeviceptr dstDevice, size_t dstPitch, unsigned char uc, size_t Width, size_t Height);
+typedef CUresult CUDAAPI tcuMemsetD2D16_v2(CUdeviceptr dstDevice, size_t dstPitch, unsigned short us, size_t Width, size_t Height);
+typedef CUresult CUDAAPI tcuMemsetD2D32_v2(CUdeviceptr dstDevice, size_t dstPitch, unsigned int ui, size_t Width, size_t Height);
+typedef CUresult CUDAAPI tcuMemsetD8Async(CUdeviceptr dstDevice, unsigned char uc, size_t N, CUstream hStream);
+typedef CUresult CUDAAPI tcuMemsetD16Async(CUdeviceptr dstDevice, unsigned short us, size_t N, CUstream hStream);
+typedef CUresult CUDAAPI tcuMemsetD32Async(CUdeviceptr dstDevice, unsigned int ui, size_t N, CUstream hStream);
+typedef CUresult CUDAAPI tcuMemsetD2D8Async(CUdeviceptr dstDevice, size_t dstPitch, unsigned char uc, size_t Width, size_t Height, CUstream hStream);
+typedef CUresult CUDAAPI tcuMemsetD2D16Async(CUdeviceptr dstDevice, size_t dstPitch, unsigned short us, size_t Width, size_t Height, CUstream hStream);
+typedef CUresult CUDAAPI tcuMemsetD2D32Async(CUdeviceptr dstDevice, size_t dstPitch, unsigned int ui, size_t Width, size_t Height, CUstream hStream);
 typedef CUresult CUDAAPI tcuArrayCreate_v2(CUarray* pHandle, const CUDA_ARRAY_DESCRIPTOR* pAllocateArray);
 typedef CUresult CUDAAPI tcuArrayGetDescriptor_v2(CUDA_ARRAY_DESCRIPTOR* pArrayDescriptor, CUarray hArray);
 typedef CUresult CUDAAPI tcuArrayDestroy(CUarray hArray);
 typedef CUresult CUDAAPI tcuArray3DCreate_v2(CUarray* pHandle, const CUDA_ARRAY3D_DESCRIPTOR* pAllocateArray);
 typedef CUresult CUDAAPI tcuArray3DGetDescriptor_v2(CUDA_ARRAY3D_DESCRIPTOR* pArrayDescriptor, CUarray hArray);
-typedef CUresult CUDAAPI tcuMipmappedArrayCreate(CUmipmappedArray* pHandle, const CUDA_ARRAY3D_DESCRIPTOR* pMipmappedArrayDesc, unsigned numMipmapLevels);
-typedef CUresult CUDAAPI tcuMipmappedArrayGetLevel(CUarray* pLevelArray, CUmipmappedArray hMipmappedArray, unsigned level);
+typedef CUresult CUDAAPI tcuMipmappedArrayCreate(CUmipmappedArray* pHandle, const CUDA_ARRAY3D_DESCRIPTOR* pMipmappedArrayDesc, unsigned int numMipmapLevels);
+typedef CUresult CUDAAPI tcuMipmappedArrayGetLevel(CUarray* pLevelArray, CUmipmappedArray hMipmappedArray, unsigned int level);
 typedef CUresult CUDAAPI tcuMipmappedArrayDestroy(CUmipmappedArray hMipmappedArray);
 typedef CUresult CUDAAPI tcuPointerGetAttribute(void* data, CUpointer_attribute attribute, CUdeviceptr ptr);
 typedef CUresult CUDAAPI tcuPointerSetAttribute(const void* value, CUpointer_attribute attribute, CUdeviceptr ptr);
-typedef CUresult CUDAAPI tcuPointerGetAttributes(unsigned numAttributes, CUpointer_attribute* attributes, void* data, CUdeviceptr ptr);
-typedef CUresult CUDAAPI tcuStreamCreate(CUstream* phStream, unsigned Flags);
-typedef CUresult CUDAAPI tcuStreamCreateWithPriority(CUstream* phStream, unsigned flags, int priority);
+typedef CUresult CUDAAPI tcuPointerGetAttributes(unsigned int numAttributes, CUpointer_attribute* attributes, void* data, CUdeviceptr ptr);
+typedef CUresult CUDAAPI tcuStreamCreate(CUstream* phStream, unsigned int Flags);
+typedef CUresult CUDAAPI tcuStreamCreateWithPriority(CUstream* phStream, unsigned int flags, int priority);
 typedef CUresult CUDAAPI tcuStreamGetPriority(CUstream hStream, int* priority);
-typedef CUresult CUDAAPI tcuStreamGetFlags(CUstream hStream, unsigned* flags);
-typedef CUresult CUDAAPI tcuStreamWaitEvent(CUstream hStream, CUevent hEvent, unsigned Flags);
-typedef CUresult CUDAAPI tcuStreamAddCallback(CUstream hStream, CUstreamCallback callback, void* userData, unsigned flags);
-typedef CUresult CUDAAPI tcuStreamAttachMemAsync(CUstream hStream, CUdeviceptr dptr, size_t length, unsigned flags);
+typedef CUresult CUDAAPI tcuStreamGetFlags(CUstream hStream, unsigned int* flags);
+typedef CUresult CUDAAPI tcuStreamWaitEvent(CUstream hStream, CUevent hEvent, unsigned int Flags);
+typedef CUresult CUDAAPI tcuStreamAddCallback(CUstream hStream, CUstreamCallback callback, void* userData, unsigned int flags);
+typedef CUresult CUDAAPI tcuStreamAttachMemAsync(CUstream hStream, CUdeviceptr dptr, size_t length, unsigned int flags);
 typedef CUresult CUDAAPI tcuStreamQuery(CUstream hStream);
 typedef CUresult CUDAAPI tcuStreamSynchronize(CUstream hStream);
 typedef CUresult CUDAAPI tcuStreamDestroy_v2(CUstream hStream);
-typedef CUresult CUDAAPI tcuEventCreate(CUevent* phEvent, unsigned Flags);
+typedef CUresult CUDAAPI tcuEventCreate(CUevent* phEvent, unsigned int Flags);
 typedef CUresult CUDAAPI tcuEventRecord(CUevent hEvent, CUstream hStream);
 typedef CUresult CUDAAPI tcuEventQuery(CUevent hEvent);
 typedef CUresult CUDAAPI tcuEventSynchronize(CUevent hEvent);
@@ -884,23 +884,23 @@ typedef CUresult CUDAAPI tcuEventElapsedTime(float* pMilliseconds, CUevent hStar
 typedef CUresult CUDAAPI tcuFuncGetAttribute(int* pi, CUfunction_attribute attrib, CUfunction hfunc);
 typedef CUresult CUDAAPI tcuFuncSetCacheConfig(CUfunction hfunc, CUfunc_cache config);
 typedef CUresult CUDAAPI tcuFuncSetSharedMemConfig(CUfunction hfunc, CUsharedconfig config);
-typedef CUresult CUDAAPI tcuLaunchKernel(CUfunction f, unsigned gridDimX, unsigned gridDimY, unsigned gridDimZ, unsigned blockDimX, unsigned blockDimY, unsigned blockDimZ, unsigned sharedMemBytes, CUstream hStream, void* kernelParams, void* extra);
+typedef CUresult CUDAAPI tcuLaunchKernel(CUfunction f, unsigned int gridDimX, unsigned int gridDimY, unsigned int gridDimZ, unsigned int blockDimX, unsigned int blockDimY, unsigned int blockDimZ, unsigned int sharedMemBytes, CUstream hStream, void* kernelParams, void* extra);
 typedef CUresult CUDAAPI tcuFuncSetBlockShape(CUfunction hfunc, int x, int y, int z);
-typedef CUresult CUDAAPI tcuFuncSetSharedSize(CUfunction hfunc, unsigned bytes);
-typedef CUresult CUDAAPI tcuParamSetSize(CUfunction hfunc, unsigned numbytes);
-typedef CUresult CUDAAPI tcuParamSeti(CUfunction hfunc, int offset, unsigned value);
+typedef CUresult CUDAAPI tcuFuncSetSharedSize(CUfunction hfunc, unsigned int bytes);
+typedef CUresult CUDAAPI tcuParamSetSize(CUfunction hfunc, unsigned int numbytes);
+typedef CUresult CUDAAPI tcuParamSeti(CUfunction hfunc, int offset, unsigned int value);
 typedef CUresult CUDAAPI tcuParamSetf(CUfunction hfunc, int offset, float value);
-typedef CUresult CUDAAPI tcuParamSetv(CUfunction hfunc, int offset, void* ptr, unsigned numbytes);
+typedef CUresult CUDAAPI tcuParamSetv(CUfunction hfunc, int offset, void* ptr, unsigned int numbytes);
 typedef CUresult CUDAAPI tcuLaunch(CUfunction f);
 typedef CUresult CUDAAPI tcuLaunchGrid(CUfunction f, int grid_width, int grid_height);
 typedef CUresult CUDAAPI tcuLaunchGridAsync(CUfunction f, int grid_width, int grid_height, CUstream hStream);
 typedef CUresult CUDAAPI tcuParamSetTexRef(CUfunction hfunc, int texunit, CUtexref hTexRef);
 typedef CUresult CUDAAPI tcuOccupancyMaxActiveBlocksPerMultiprocessor(int* numBlocks, CUfunction func, int blockSize, size_t dynamicSMemSize);
-typedef CUresult CUDAAPI tcuOccupancyMaxActiveBlocksPerMultiprocessorWithFlags(int* numBlocks, CUfunction func, int blockSize, size_t dynamicSMemSize, unsigned flags);
+typedef CUresult CUDAAPI tcuOccupancyMaxActiveBlocksPerMultiprocessorWithFlags(int* numBlocks, CUfunction func, int blockSize, size_t dynamicSMemSize, unsigned int flags);
 typedef CUresult CUDAAPI tcuOccupancyMaxPotentialBlockSize(int* minGridSize, int* blockSize, CUfunction func, CUoccupancyB2DSize blockSizeToDynamicSMemSize, size_t dynamicSMemSize, int blockSizeLimit);
-typedef CUresult CUDAAPI tcuOccupancyMaxPotentialBlockSizeWithFlags(int* minGridSize, int* blockSize, CUfunction func, CUoccupancyB2DSize blockSizeToDynamicSMemSize, size_t dynamicSMemSize, int blockSizeLimit, unsigned flags);
-typedef CUresult CUDAAPI tcuTexRefSetArray(CUtexref hTexRef, CUarray hArray, unsigned Flags);
-typedef CUresult CUDAAPI tcuTexRefSetMipmappedArray(CUtexref hTexRef, CUmipmappedArray hMipmappedArray, unsigned Flags);
+typedef CUresult CUDAAPI tcuOccupancyMaxPotentialBlockSizeWithFlags(int* minGridSize, int* blockSize, CUfunction func, CUoccupancyB2DSize blockSizeToDynamicSMemSize, size_t dynamicSMemSize, int blockSizeLimit, unsigned int flags);
+typedef CUresult CUDAAPI tcuTexRefSetArray(CUtexref hTexRef, CUarray hArray, unsigned int Flags);
+typedef CUresult CUDAAPI tcuTexRefSetMipmappedArray(CUtexref hTexRef, CUmipmappedArray hMipmappedArray, unsigned int Flags);
 typedef CUresult CUDAAPI tcuTexRefSetAddress_v2(size_t* ByteOffset, CUtexref hTexRef, CUdeviceptr dptr, size_t bytes);
 typedef CUresult CUDAAPI tcuTexRefSetAddress2D_v3(CUtexref hTexRef, const CUDA_ARRAY_DESCRIPTOR* desc, CUdeviceptr dptr, size_t Pitch);
 typedef CUresult CUDAAPI tcuTexRefSetFormat(CUtexref hTexRef, CUarray_format fmt, int NumPackedComponents);
@@ -909,8 +909,8 @@ typedef CUresult CUDAAPI tcuTexRefSetFilterMode(CUtexref hTexRef, CUfilter_mode
 typedef CUresult CUDAAPI tcuTexRefSetMipmapFilterMode(CUtexref hTexRef, CUfilter_mode fm);
 typedef CUresult CUDAAPI tcuTexRefSetMipmapLevelBias(CUtexref hTexRef, float bias);
 typedef CUresult CUDAAPI tcuTexRefSetMipmapLevelClamp(CUtexref hTexRef, float minMipmapLevelClamp, float maxMipmapLevelClamp);
-typedef CUresult CUDAAPI tcuTexRefSetMaxAnisotropy(CUtexref hTexRef, unsigned maxAniso);
-typedef CUresult CUDAAPI tcuTexRefSetFlags(CUtexref hTexRef, unsigned Flags);
+typedef CUresult CUDAAPI tcuTexRefSetMaxAnisotropy(CUtexref hTexRef, unsigned int maxAniso);
+typedef CUresult CUDAAPI tcuTexRefSetFlags(CUtexref hTexRef, unsigned int Flags);
 typedef CUresult CUDAAPI tcuTexRefGetAddress_v2(CUdeviceptr* pdptr, CUtexref hTexRef);
 typedef CUresult CUDAAPI tcuTexRefGetArray(CUarray* phArray, CUtexref hTexRef);
 typedef CUresult CUDAAPI tcuTexRefGetMipmappedArray(CUmipmappedArray* phMipmappedArray, CUtexref hTexRef);
@@ -921,10 +921,10 @@ typedef CUresult CUDAAPI tcuTexRefGetMipmapFilterMode(CUfilter_mode* pfm, CUtexr
 typedef CUresult CUDAAPI tcuTexRefGetMipmapLevelBias(float* pbias, CUtexref hTexRef);
 typedef CUresult CUDAAPI tcuTexRefGetMipmapLevelClamp(float* pminMipmapLevelClamp, float* pmaxMipmapLevelClamp, CUtexref hTexRef);
 typedef CUresult CUDAAPI tcuTexRefGetMaxAnisotropy(int* pmaxAniso, CUtexref hTexRef);
-typedef CUresult CUDAAPI tcuTexRefGetFlags(unsigned* pFlags, CUtexref hTexRef);
+typedef CUresult CUDAAPI tcuTexRefGetFlags(unsigned int* pFlags, CUtexref hTexRef);
 typedef CUresult CUDAAPI tcuTexRefCreate(CUtexref* pTexRef);
 typedef CUresult CUDAAPI tcuTexRefDestroy(CUtexref hTexRef);
-typedef CUresult CUDAAPI tcuSurfRefSetArray(CUsurfref hSurfRef, CUarray hArray, unsigned Flags);
+typedef CUresult CUDAAPI tcuSurfRefSetArray(CUsurfref hSurfRef, CUarray hArray, unsigned int Flags);
 typedef CUresult CUDAAPI tcuSurfRefGetArray(CUarray* phArray, CUsurfref hSurfRef);
 typedef CUresult CUDAAPI tcuTexObjectCreate(CUtexObject* pTexObject, const CUDA_RESOURCE_DESC* pResDesc, const CUDA_TEXTURE_DESC* pTexDesc, const CUDA_RESOURCE_VIEW_DESC* pResViewDesc);
 typedef CUresult CUDAAPI tcuTexObjectDestroy(CUtexObject texObject);
@@ -935,27 +935,27 @@ typedef CUresult CUDAAPI tcuSurfObjectCreate(CUsurfObject* pSurfObject, const CU
 typedef CUresult CUDAAPI tcuSurfObjectDestroy(CUsurfObject surfObject);
 typedef CUresult CUDAAPI tcuSurfObjectGetResourceDesc(CUDA_RESOURCE_DESC* pResDesc, CUsurfObject surfObject);
 typedef CUresult CUDAAPI tcuDeviceCanAccessPeer(int* canAccessPeer, CUdevice dev, CUdevice peerDev);
-typedef CUresult CUDAAPI tcuCtxEnablePeerAccess(CUcontext peerContext, unsigned Flags);
+typedef CUresult CUDAAPI tcuCtxEnablePeerAccess(CUcontext peerContext, unsigned int Flags);
 typedef CUresult CUDAAPI tcuCtxDisablePeerAccess(CUcontext peerContext);
 typedef CUresult CUDAAPI tcuGraphicsUnregisterResource(CUgraphicsResource resource);
-typedef CUresult CUDAAPI tcuGraphicsSubResourceGetMappedArray(CUarray* pArray, CUgraphicsResource resource, unsigned arrayIndex, unsigned mipLevel);
+typedef CUresult CUDAAPI tcuGraphicsSubResourceGetMappedArray(CUarray* pArray, CUgraphicsResource resource, unsigned int arrayIndex, unsigned int mipLevel);
 typedef CUresult CUDAAPI tcuGraphicsResourceGetMappedMipmappedArray(CUmipmappedArray* pMipmappedArray, CUgraphicsResource resource);
 typedef CUresult CUDAAPI tcuGraphicsResourceGetMappedPointer_v2(CUdeviceptr* pDevPtr, size_t* pSize, CUgraphicsResource resource);
-typedef CUresult CUDAAPI tcuGraphicsResourceSetMapFlags_v2(CUgraphicsResource resource, unsigned flags);
-typedef CUresult CUDAAPI tcuGraphicsMapResources(unsigned count, CUgraphicsResource* resources, CUstream hStream);
-typedef CUresult CUDAAPI tcuGraphicsUnmapResources(unsigned count, CUgraphicsResource* resources, CUstream hStream);
+typedef CUresult CUDAAPI tcuGraphicsResourceSetMapFlags_v2(CUgraphicsResource resource, unsigned int flags);
+typedef CUresult CUDAAPI tcuGraphicsMapResources(unsigned int count, CUgraphicsResource* resources, CUstream hStream);
+typedef CUresult CUDAAPI tcuGraphicsUnmapResources(unsigned int count, CUgraphicsResource* resources, CUstream hStream);
 typedef CUresult CUDAAPI tcuGetExportTable(const void* ppExportTable, const CUuuid* pExportTableId);
 
-typedef CUresult CUDAAPI tcuGraphicsGLRegisterBuffer(CUgraphicsResource* pCudaResource, GLuint buffer, unsigned Flags);
-typedef CUresult CUDAAPI tcuGraphicsGLRegisterImage(CUgraphicsResource* pCudaResource, GLuint image, GLenum target, unsigned Flags);
-typedef CUresult CUDAAPI tcuGLGetDevices_v2(unsigned* pCudaDeviceCount, CUdevice* pCudaDevices, unsigned cudaDeviceCount, CUGLDeviceList deviceList);
-typedef CUresult CUDAAPI tcuGLCtxCreate_v2(CUcontext* pCtx, unsigned Flags, CUdevice device);
+typedef CUresult CUDAAPI tcuGraphicsGLRegisterBuffer(CUgraphicsResource* pCudaResource, GLuint buffer, unsigned int Flags);
+typedef CUresult CUDAAPI tcuGraphicsGLRegisterImage(CUgraphicsResource* pCudaResource, GLuint image, GLenum target, unsigned int Flags);
+typedef CUresult CUDAAPI tcuGLGetDevices_v2(unsigned int* pCudaDeviceCount, CUdevice* pCudaDevices, unsigned int cudaDeviceCount, CUGLDeviceList deviceList);
+typedef CUresult CUDAAPI tcuGLCtxCreate_v2(CUcontext* pCtx, unsigned int Flags, CUdevice device);
 typedef CUresult CUDAAPI tcuGLInit(void);
 typedef CUresult CUDAAPI tcuGLRegisterBufferObject(GLuint buffer);
 typedef CUresult CUDAAPI tcuGLMapBufferObject_v2(CUdeviceptr* dptr, size_t* size, GLuint buffer);
 typedef CUresult CUDAAPI tcuGLUnmapBufferObject(GLuint buffer);
 typedef CUresult CUDAAPI tcuGLUnregisterBufferObject(GLuint buffer);
-typedef CUresult CUDAAPI tcuGLSetBufferObjectMapFlags(GLuint buffer, unsigned Flags);
+typedef CUresult CUDAAPI tcuGLSetBufferObjectMapFlags(GLuint buffer, unsigned int Flags);
 typedef CUresult CUDAAPI tcuGLMapBufferObjectAsync_v2(CUdeviceptr* dptr, size_t* size, GLuint buffer, CUstream hStream);
 typedef CUresult CUDAAPI tcuGLUnmapBufferObjectAsync(GLuint buffer, CUstream hStream);
 

extern/curve_fit_nd/CMakeLists.txt

@@ -0,0 +1,35 @@
+# ***** 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.
+#
+# ***** END GPL LICENSE BLOCK *****
+
+set(INC
+	.
+)
+
+set(INC_SYS
+
+)
+
+set(SRC
+	intern/curve_fit_cubic.c
+	intern/curve_fit_corners_detect.c
+
+	intern/curve_fit_inline.h
+	curve_fit_nd.h
+)
+
+blender_add_lib(extern_curve_fit_nd "${SRC}" "${INC}" "${INC_SYS}")

extern/curve_fit_nd/curve_fit_nd.h

@@ -0,0 +1,162 @@
+/*
+ * Copyright (c) 2016, DWANGO Co., Ltd.
+ * All rights reserved.
+ *
+ * Redistribution and use in source and binary forms, with or without
+ * modification, are permitted provided that the following conditions are met:
+ *     * Redistributions of source code must retain the above copyright
+ *       notice, this list of conditions and the following disclaimer.
+ *     * Redistributions in binary form must reproduce the above copyright
+ *       notice, this list of conditions and the following disclaimer in the
+ *       documentation and/or other materials provided with the distribution.
+ *     * Neither the name of the <organization> nor the
+ *       names of its contributors may be used to endorse or promote products
+ *       derived from this software without specific prior written permission.
+ *
+ * THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS "AS IS" AND
+ * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE IMPLIED
+ * WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE ARE
+ * DISCLAIMED. IN NO EVENT SHALL <COPYRIGHT HOLDER> BE LIABLE FOR ANY
+ * DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES
+ * (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES;
+ * LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND
+ * ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT
+ * (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE OF THIS
+ * SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
+ */
+
+#ifndef __CURVE_FIT_ND_H__
+#define __CURVE_FIT_ND_H__
+
+/** \file curve_fit_nd.h
+ *  \ingroup curve_fit
+ */
+
+
+/* curve_fit_cubic.c */
+
+/**
+ * Takes a flat array of points and evalues that to calculate a bezier spline.
+ *
+ * \param points, points_len: The array of points to calculate a cubics from.
+ * \param dims: The number of dimensions for for each element in \a points.
+ * \param error_threshold: the error threshold to allow for,
+ * the curve will be within this distance from \a points.
+ * \param corners, corners_len: indices for points which will not have aligned tangents (optional).
+ * This can use the output of #curve_fit_corners_detect_db which has been included
+ * to evaluate a line to detect corner indices.
+ *
+ * \param r_cubic_array, r_cubic_array_len: Resulting array of tangents and knots, formatted as follows:
+ * ``r_cubic_array[r_cubic_array_len][3][dims]``,
+ * where each point has 0 and 2 for the tangents and the middle index 1 for the knot.
+ * The size of the *flat* array will be ``r_cubic_array_len * 3 * dims``.
+ * \param r_corner_index_array, r_corner_index_len: Corner indices in in \a r_cubic_array (optional).
+ * This allows you to access corners on the resulting curve.
+ *
+ * \returns zero on success, nonzero is reserved for error values.
+ */
+int curve_fit_cubic_to_points_db(
+        const double       *points,
+        const unsigned int  points_len,
+        const unsigned int  dims,
+        const double        error_threshold,
+        const unsigned int *corners,
+        unsigned int        corners_len,
+
+        double **r_cubic_array, unsigned int *r_cubic_array_len,
+        unsigned int **r_cubic_orig_index,
+        unsigned int **r_corner_index_array, unsigned int *r_corner_index_len);
+
+int curve_fit_cubic_to_points_fl(
+        const float        *points,
+        const unsigned int  points_len,
+        const unsigned int  dims,
+        const float         error_threshold,
+        const unsigned int *corners,
+        const unsigned int  corners_len,
+
+        float **r_cubic_array, unsigned int *r_cubic_array_len,
+        unsigned int **r_cubic_orig_index,
+        unsigned int **r_corners_index_array, unsigned int *r_corners_index_len);
+
+/**
+ * Takes a flat array of points and evalues that to calculate handle lengths.
+ *
+ * \param points, points_len: The array of points to calculate a cubics from.
+ * \param dims: The number of dimensions for for each element in \a points.
+ * \param error_threshold: the error threshold to allow for,
+ * \param tan_l, tan_r: Normalized tangents the handles will be aligned to.
+ * Note that tangents must both point along the direction of the \a points,
+ * so \a tan_l points in the same direction of the resulting handle,
+ * where \a tan_r will point the opposite direction of its handle.
+ *
+ * \param r_handle_l, r_handle_r: Resulting calculated handles.
+ * \param r_error_sq: The maximum distance  (squared) this curve diverges from \a points.
+ */
+int curve_fit_cubic_to_points_single_db(
+        const double      *points,
+        const unsigned int points_len,
+        const unsigned int dims,
+        const double       error_threshold,
+        const double       tan_l[],
+        const double       tan_r[],
+
+        double  r_handle_l[],
+        double  r_handle_r[],
+        double *r_error_sq);
+
+int curve_fit_cubic_to_points_single_fl(
+        const float       *points,
+        const unsigned int points_len,
+        const unsigned int dims,
+        const float        error_threshold,
+        const float        tan_l[],
+        const float        tan_r[],
+
+        float   r_handle_l[],
+        float   r_handle_r[],
+        float  *r_error_sq);
+
+/* curve_fit_corners_detect.c */
+
+/**
+ * A helper function that takes a line and outputs its corner indices.
+ *
+ * \param points, points_len: Curve to evaluate.
+ * \param dims: The number of dimensions for for each element in \a points.
+ * \param radius_min: Corners on the curve between points below this radius are ignored.
+ * \param radius_max: Corners on the curve above this radius are ignored.
+ * \param samples_max: Prevent testing corners beyond this many points
+ * (prevents a large radius taking excessive time to compute).
+ * \param angle_threshold: Angles above this value are considered corners
+ * (higher value for fewer corners).
+ *
+ * \param r_corners, r_corners_len: Resulting array of corners.
+ *
+ * \returns zero on success, nonzero is reserved for error values.
+ */
+int curve_fit_corners_detect_db(
+        const double      *points,
+        const unsigned int points_len,
+        const unsigned int dims,
+        const double       radius_min,
+        const double       radius_max,
+        const unsigned int samples_max,
+        const double       angle_threshold,
+
+        unsigned int **r_corners,
+        unsigned int  *r_corners_len);
+
+int curve_fit_corners_detect_fl(
+        const float       *points,
+        const unsigned int points_len,
+        const unsigned int dims,
+        const float        radius_min,
+        const float        radius_max,
+        const unsigned int samples_max,
+        const float        angle_threshold,
+
+        unsigned int **r_corners,
+        unsigned int  *r_corners_len);
+
+#endif  /* __CURVE_FIT_ND_H__ */

extern/curve_fit_nd/intern/curve_fit_corners_detect.c

@@ -0,0 +1,468 @@
+/*
+ * Copyright (c) 2016, Blender Foundation.
+ * All rights reserved.
+ *
+ * Redistribution and use in source and binary forms, with or without
+ * modification, are permitted provided that the following conditions are met:
+ *     * Redistributions of source code must retain the above copyright
+ *       notice, this list of conditions and the following disclaimer.
+ *     * Redistributions in binary form must reproduce the above copyright
+ *       notice, this list of conditions and the following disclaimer in the
+ *       documentation and/or other materials provided with the distribution.
+ *     * Neither the name of the <organization> nor the
+ *       names of its contributors may be used to endorse or promote products
+ *       derived from this software without specific prior written permission.
+ *
+ * THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS "AS IS" AND
+ * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE IMPLIED
+ * WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE ARE
+ * DISCLAIMED. IN NO EVENT SHALL <COPYRIGHT HOLDER> BE LIABLE FOR ANY
+ * DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES
+ * (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES;
+ * LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND
+ * ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT
+ * (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE OF THIS
+ * SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
+ */
+
+/** \file curve_fit_corners_detect.c
+ *  \ingroup curve_fit
+ */
+
+#include <math.h>
+#include <float.h>
+#include <stdbool.h>
+#include <assert.h>
+
+#include <string.h>
+#include <stdlib.h>
+
+#include "../curve_fit_nd.h"
+
+typedef unsigned int uint;
+
+#include "curve_fit_inline.h"
+
+#ifdef _MSC_VER
+#  define alloca(size) _alloca(size)
+#endif
+
+#if !defined(_MSC_VER)
+#  define USE_VLA
+#endif
+
+#ifdef USE_VLA
+#  ifdef __GNUC__
+#    pragma GCC diagnostic ignored "-Wvla"
+#  endif
+#else
+#  ifdef __GNUC__
+#    pragma GCC diagnostic error "-Wvla"
+#  endif
+#endif
+
+/* -------------------------------------------------------------------- */
+
+/** \name Simple Vector Math Lib
+ * \{ */
+
+static double cos_vnvnvn(
+        const double v0[], const double v1[], const double v2[],
+        const uint dims)
+{
+#ifdef USE_VLA
+	double dvec0[dims];
+	double dvec1[dims];
+#else
+	double *dvec0 = alloca(sizeof(double) * dims);
+	double *dvec1 = alloca(sizeof(double) * dims);
+#endif
+	normalize_vn_vnvn(dvec0, v0, v1, dims);
+	normalize_vn_vnvn(dvec1, v1, v2, dims);
+	double d = dot_vnvn(dvec0, dvec1, dims);
+	/* sanity check */
+	d = max(-1.0, min(1.0, d));
+	return d;
+}
+
+static double angle_vnvnvn(
+        const double v0[], const double v1[], const double v2[],
+        const uint dims)
+{
+	return acos(cos_vnvnvn(v0, v1, v2, dims));
+}
+
+
+static bool isect_line_sphere_vn(
+        const double l1[],
+        const double l2[],
+        const double sp[],
+        const double r,
+        uint dims,
+
+        double r_p1[]
+#if 0   /* UNUSED */
+        double r_p2[]
+#endif
+        )
+{
+#ifdef USE_VLA
+	double ldir[dims];
+	double tvec[dims];
+#else
+	double *ldir = alloca(sizeof(double) * dims);
+	double *tvec = alloca(sizeof(double) * dims);
+#endif
+
+	sub_vn_vnvn(ldir, l2, l1, dims);
+
+	sub_vn_vnvn(tvec, l1, sp, dims);
+	const double a = len_squared_vn(ldir, dims);
+	const double b = 2.0 * dot_vnvn(ldir, tvec, dims);
+	const double c = len_squared_vn(sp, dims) + len_squared_vn(l1, dims) - (2.0 * dot_vnvn(sp, l1, dims)) - sq(r);
+
+	const double i = b * b - 4.0 * a * c;
+
+	if ((i < 0.0) || (a == 0.0)) {
+		return false;
+	}
+	else if (i == 0.0) {
+		/* one intersection */
+		const double mu = -b / (2.0 * a);
+		mul_vnvn_fl(r_p1, ldir, mu, dims);
+		iadd_vnvn(r_p1, l1, dims);
+		return true;
+	}
+	else if (i > 0.0) {
+		/* # avoid calc twice */
+		const double i_sqrt = sqrt(i);
+		double mu;
+
+		/* Note: when l1 is inside the sphere and l2 is outside.
+		 * the first intersection point will always be between the pair. */
+
+		/* first intersection */
+		mu = (-b + i_sqrt) / (2.0 * a);
+		mul_vnvn_fl(r_p1, ldir, mu, dims);
+		iadd_vnvn(r_p1, l1, dims);
+#if 0
+		/* second intersection */
+		mu = (-b - i_sqrt) / (2.0 * a);
+		mul_vnvn_fl(r_p2, ldir, mu, dims);
+		iadd_vnvn(r_p2, l1, dims);
+#endif
+		return true;
+	}
+	else {
+		return false;
+	}
+}
+
+/** \} */
+
+
+/* -------------------------------------------------------------------- */
+
+
+static bool point_corner_measure(
+        const double *points,
+        const uint    points_len,
+        const uint i,
+        const uint i_prev_init,
+        const uint i_next_init,
+        const double radius,
+        const uint samples_max,
+        const uint dims,
+
+        double r_p_prev[], uint *r_i_prev_next,
+        double r_p_next[], uint *r_i_next_prev)
+{
+	const double *p = &points[i * dims];
+	uint sample;
+
+
+	uint i_prev = i_prev_init;
+	uint i_prev_next = i_prev + 1;
+	sample = 0;
+	while (true) {
+		if ((i_prev == -1) || (sample++ > samples_max)) {
+			return false;
+		}
+		else if (len_squared_vnvn(p, &points[i_prev * dims], dims) < radius) {
+			i_prev -= 1;
+		}
+		else {
+			break;
+		}
+	}
+
+	uint i_next = i_next_init;
+	uint i_next_prev = i_next - 1;
+	sample = 0;
+	while (true) {
+		if ((i_next == points_len) || (sample++ > samples_max)) {
+			return false;
+		}
+		else if (len_squared_vnvn(p, &points[i_next * dims], dims) < radius) {
+			i_next += 1;
+		}
+		else {
+			break;
+		}
+	}
+
+	/* find points on the sphere */
+	if (!isect_line_sphere_vn(
+	        &points[i_prev * dims], &points[i_prev_next * dims], p, radius, dims,
+	        r_p_prev))
+	{
+		return false;
+	}
+
+	if (!isect_line_sphere_vn(
+	        &points[i_next * dims], &points[i_next_prev * dims], p, radius, dims,
+	        r_p_next))
+	{
+		return false;
+	}
+
+	*r_i_prev_next = i_prev_next;
+	*r_i_next_prev = i_next_prev;
+
+	return true;
+}
+
+
+static double point_corner_angle(
+        const double *points,
+        const uint    points_len,
+        const uint i,
+        const double radius_mid,
+        const double radius_max,
+        const double angle_threshold,
+        const double angle_threshold_cos,
+        /* prevent locking up when for example `radius_min` is very large
+         * (possibly larger then the curve).
+         * In this case we would end up checking every point from every other point,
+         * never reaching one that was outside the `radius_min`. */
+
+        /* prevent locking up when for e */
+        const uint samples_max,
+
+        const uint dims)
+{
+	assert(angle_threshold_cos == cos(angle_threshold));
+
+	if (i == 0 || i == points_len - 1) {
+		return 0.0;
+	}
+
+	const double *p = &points[i * dims];
+
+	/* initial test */
+	if (cos_vnvnvn(&points[(i - 1) * dims], p, &points[(i + 1) * dims], dims) > angle_threshold_cos) {
+		return 0.0;
+	}
+
+#ifdef USE_VLA
+	double p_mid_prev[dims];
+	double p_mid_next[dims];
+#else
+	double *p_mid_prev = alloca(sizeof(double) * dims);
+	double *p_mid_next = alloca(sizeof(double) * dims);
+#endif
+
+	uint i_mid_prev_next, i_mid_next_prev;
+	if (point_corner_measure(
+	        points, points_len,
+	        i, i - 1, i + 1,
+	        radius_mid,
+	        samples_max,
+	        dims,
+
+	        p_mid_prev, &i_mid_prev_next,
+	        p_mid_next, &i_mid_next_prev))
+	{
+		const double angle_mid_cos = cos_vnvnvn(p_mid_prev, p, p_mid_next, dims);
+
+		/* compare as cos and flip direction */
+
+		/* if (angle_mid > angle_threshold) { */
+		if (angle_mid_cos < angle_threshold_cos) {
+#ifdef USE_VLA
+			double p_max_prev[dims];
+			double p_max_next[dims];
+#else
+			double *p_max_prev = alloca(sizeof(double) * dims);
+			double *p_max_next = alloca(sizeof(double) * dims);
+#endif
+
+			uint i_max_prev_next, i_max_next_prev;
+			if (point_corner_measure(
+			        points, points_len,
+			        i, i - 1, i + 1,
+			        radius_max,
+			        samples_max,
+			        dims,
+
+			        p_max_prev, &i_max_prev_next,
+			        p_max_next, &i_max_next_prev))
+			{
+				const double angle_mid = acos(angle_mid_cos);
+				const double angle_max = angle_vnvnvn(p_max_prev, p, p_max_next, dims) / 2.0;
+				const double angle_diff = angle_mid - angle_max;
+				if (angle_diff > angle_threshold) {
+					return angle_diff;
+				}
+			}
+		}
+	}
+
+	return 0.0;
+}
+
+
+int curve_fit_corners_detect_db(
+        const double *points,
+        const uint    points_len,
+        const uint dims,
+        const double radius_min,  /* ignore values below this */
+        const double radius_max,  /* ignore values above this */
+        const uint samples_max,
+        const double angle_threshold,
+
+        uint **r_corners,
+        uint  *r_corners_len)
+{
+	const double angle_threshold_cos = cos(angle_threshold);
+	uint corners_len = 0;
+
+	/* Use the difference in angle between the mid-max radii
+	 * to detect the difference between a corner and a sharp turn. */
+	const double radius_mid = (radius_min + radius_max) / 2.0;
+
+	/* we could ignore first/last- but simple to keep aligned with the point array */
+	double *points_angle = malloc(sizeof(double) * points_len);
+	points_angle[0] = 0.0;
+
+	*r_corners = NULL;
+	*r_corners_len = 0;
+
+	for (uint i = 0; i < points_len; i++) {
+		points_angle[i] =  point_corner_angle(
+		        points, points_len, i,
+		        radius_mid, radius_max,
+		        angle_threshold, angle_threshold_cos,
+		        samples_max,
+		        dims);
+
+		if (points_angle[i] != 0.0) {
+			corners_len++;
+		}
+	}
+
+	if (corners_len == 0) {
+		free(points_angle);
+		return 0;
+	}
+
+	/* Clean angle limits!
+	 *
+	 * How this works:
+	 * - Find contiguous 'corners' (where the distance is less or equal to the error threshold).
+	 * - Keep track of the corner with the highest angle
+	 * - Clear every other angle (so they're ignored when setting corners). */
+	{
+		const double radius_min_sq = sq(radius_min);
+		uint i_span_start = 0;
+		while (i_span_start < points_len) {
+			uint i_span_end = i_span_start;
+			if (points_angle[i_span_start] != 0.0) {
+				uint i_next = i_span_start + 1;
+				uint i_best = i_span_start;
+				while (i_next < points_len) {
+					if ((points_angle[i_next] == 0.0) ||
+					    (len_squared_vnvn(
+					         &points[(i_next - 1) * dims],
+					         &points[i_next * dims], dims) > radius_min_sq))
+					{
+						break;
+					}
+					else {
+						if (points_angle[i_best] < points_angle[i_next]) {
+							i_best = i_next;
+						}
+						i_span_end = i_next;
+						i_next += 1;
+					}
+				}
+
+				if (i_span_start != i_span_end) {
+					uint i = i_span_start;
+					while (i <= i_span_end) {
+						if (i != i_best) {
+							/* we could use some other error code */
+							assert(points_angle[i] != 0.0);
+							points_angle[i] = 0.0;
+							corners_len--;
+						}
+						i += 1;
+					}
+				}
+			}
+			i_span_start = i_span_end + 1;
+		}
+	}
+	/* End angle limit cleaning! */
+
+	corners_len += 2;  /* first and last */
+	uint *corners = malloc(sizeof(uint) * corners_len);
+	uint i_corner = 0;
+	corners[i_corner++] = 0;
+	for (uint i = 0; i < points_len; i++) {
+		if (points_angle[i] != 0.0) {
+			corners[i_corner++] = i;
+		}
+	}
+	corners[i_corner++] = points_len - 1;
+	assert(i_corner == corners_len);
+
+	free(points_angle);
+
+	*r_corners = corners;
+	*r_corners_len = corners_len;
+
+	return 0;
+}
+
+int curve_fit_corners_detect_fl(
+        const float *points,
+        const uint   points_len,
+        const uint dims,
+        const float radius_min,  /* ignore values below this */
+        const float radius_max,  /* ignore values above this */
+        const uint samples_max,
+        const float angle_threshold,
+
+        uint **r_corners,
+        uint  *r_corners_len)
+{
+	const uint points_flat_len = points_len * dims;
+	double *points_db = malloc(sizeof(double) * points_flat_len);
+
+	for (uint i = 0; i < points_flat_len; i++) {
+		points_db[i] = (double)points[i];
+	}
+
+	int result = curve_fit_corners_detect_db(
+	        points_db, points_len,
+	        dims,
+	        radius_min, radius_max,
+	        samples_max,
+	        angle_threshold,
+	        r_corners, r_corners_len);
+
+	free(points_db);
+
+	return result;
+}

extern/curve_fit_nd/intern/curve_fit_inline.h

@@ -0,0 +1,293 @@
+/*
+ * Copyright (c) 2016, Blender Foundation.
+ * All rights reserved.
+ *
+ * Redistribution and use in source and binary forms, with or without
+ * modification, are permitted provided that the following conditions are met:
+ *     * Redistributions of source code must retain the above copyright
+ *       notice, this list of conditions and the following disclaimer.
+ *     * Redistributions in binary form must reproduce the above copyright
+ *       notice, this list of conditions and the following disclaimer in the
+ *       documentation and/or other materials provided with the distribution.
+ *     * Neither the name of the <organization> nor the
+ *       names of its contributors may be used to endorse or promote products
+ *       derived from this software without specific prior written permission.
+ *
+ * THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS "AS IS" AND
+ * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE IMPLIED
+ * WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE ARE
+ * DISCLAIMED. IN NO EVENT SHALL <COPYRIGHT HOLDER> BE LIABLE FOR ANY
+ * DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES
+ * (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES;
+ * LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND
+ * ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT
+ * (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE OF THIS
+ * SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
+ */
+
+
+/** \file curve_fit_inline.h
+ *  \ingroup curve_fit
+ */
+
+/** \name Simple Vector Math Lib
+ * \{ */
+
+#ifdef _MSC_VER
+#  define MINLINE static __forceinline
+#else
+#  define MINLINE static inline
+#endif
+
+MINLINE double sq(const double d)
+{
+	return d * d;
+}
+
+#ifndef _MSC_VER
+MINLINE double min(const double a, const double b)
+{
+	return b < a ? b : a;
+}
+
+MINLINE double max(const double a, const double b)
+{
+	return a < b ? b : a;
+}
+#endif
+
+MINLINE void zero_vn(
+        double v0[], const uint dims)
+{
+	for (uint j = 0; j < dims; j++) {
+		v0[j] = 0.0;
+	}
+}
+
+MINLINE void flip_vn_vnvn(
+        double v_out[], const double v0[], const double v1[], const uint dims)
+{
+	for (uint j = 0; j < dims; j++) {
+		v_out[j] = v0[j] + (v0[j] - v1[j]);
+	}
+}
+
+MINLINE void copy_vnvn(
+        double v0[], const double v1[], const uint dims)
+{
+	for (uint j = 0; j < dims; j++) {
+		v0[j] = v1[j];
+	}
+}
+
+MINLINE void copy_vnfl_vndb(
+        float v0[], const double v1[], const uint dims)
+{
+	for (uint j = 0; j < dims; j++) {
+		v0[j] = (float)v1[j];
+	}
+}
+
+MINLINE void copy_vndb_vnfl(
+        double v0[], const float v1[], const uint dims)
+{
+	for (uint j = 0; j < dims; j++) {
+		v0[j] = (double)v1[j];
+	}
+}
+
+MINLINE double dot_vnvn(
+        const double v0[], const double v1[], const uint dims)
+{
+	double d = 0.0;
+	for (uint j = 0; j < dims; j++) {
+		d += v0[j] * v1[j];
+	}
+	return d;
+}
+
+MINLINE void add_vn_vnvn(
+        double v_out[], const double v0[], const double v1[], const uint dims)
+{
+	for (uint j = 0; j < dims; j++) {
+		v_out[j] = v0[j] + v1[j];
+	}
+}
+
+MINLINE void sub_vn_vnvn(
+        double v_out[], const double v0[], const double v1[], const uint dims)
+{
+	for (uint j = 0; j < dims; j++) {
+		v_out[j] = v0[j] - v1[j];
+	}
+}
+
+MINLINE void iadd_vnvn(
+        double v0[], const double v1[], const uint dims)
+{
+	for (uint j = 0; j < dims; j++) {
+		v0[j] += v1[j];
+	}
+}
+
+MINLINE void isub_vnvn(
+        double v0[], const double v1[], const uint dims)
+{
+	for (uint j = 0; j < dims; j++) {
+		v0[j] -= v1[j];
+	}
+}
+
+MINLINE void madd_vn_vnvn_fl(
+        double v_out[],
+        const double v0[], const double v1[],
+        const double f, const uint dims)
+{
+	for (uint j = 0; j < dims; j++) {
+		v_out[j] = v0[j] + v1[j] * f;
+	}
+}
+
+MINLINE void msub_vn_vnvn_fl(
+        double v_out[],
+        const double v0[], const double v1[],
+        const double f, const uint dims)
+{
+	for (uint j = 0; j < dims; j++) {
+		v_out[j] = v0[j] - v1[j] * f;
+	}
+}
+
+MINLINE void miadd_vn_vn_fl(
+        double v_out[], const double v0[], double f, const uint dims)
+{
+	for (uint j = 0; j < dims; j++) {
+		v_out[j] += v0[j] * f;
+	}
+}
+
+#if 0
+MINLINE void misub_vn_vn_fl(
+        double v_out[], const double v0[], double f, const uint dims)
+{
+	for (uint j = 0; j < dims; j++) {
+		v_out[j] -= v0[j] * f;
+	}
+}
+#endif
+
+MINLINE void mul_vnvn_fl(
+        double v_out[],
+        const double v0[], const double f, const uint dims)
+{
+	for (uint j = 0; j < dims; j++) {
+		v_out[j] = v0[j] * f;
+	}
+}
+
+MINLINE void imul_vn_fl(double v0[], const double f, const uint dims)
+{
+	for (uint j = 0; j < dims; j++) {
+		v0[j] *= f;
+	}
+}
+
+
+MINLINE double len_squared_vnvn(
+        const double v0[], const double v1[], const uint dims)
+{
+	double d = 0.0;
+	for (uint j = 0; j < dims; j++) {
+		d += sq(v0[j] - v1[j]);
+	}
+	return d;
+}
+
+MINLINE double len_squared_vn(
+        const double v0[], const uint dims)
+{
+	double d = 0.0;
+	for (uint j = 0; j < dims; j++) {
+		d += sq(v0[j]);
+	}
+	return d;
+}
+
+MINLINE double len_vnvn(
+        const double v0[], const double v1[], const uint dims)
+{
+	return sqrt(len_squared_vnvn(v0, v1, dims));
+}
+
+MINLINE double len_vn(
+        const double v0[], const uint dims)
+{
+	return sqrt(len_squared_vn(v0, dims));
+}
+
+/* special case, save us negating a copy, then getting the length */
+MINLINE double len_squared_negated_vnvn(
+        const double v0[], const double v1[], const uint dims)
+{
+	double d = 0.0;
+	for (uint j = 0; j < dims; j++) {
+		d += sq(v0[j] + v1[j]);
+	}
+	return d;
+}
+
+MINLINE double len_negated_vnvn(
+        const double v0[], const double v1[], const uint dims)
+{
+	return sqrt(len_squared_negated_vnvn(v0, v1, dims));
+}
+
+MINLINE double normalize_vn(
+        double v0[], const uint dims)
+{
+	double d = len_squared_vn(v0, dims);
+	if (d != 0.0 && ((d = sqrt(d)) != 0.0)) {
+		imul_vn_fl(v0, 1.0 / d, dims);
+	}
+	return d;
+}
+
+/* v_out = (v0 - v1).normalized() */
+MINLINE double normalize_vn_vnvn(
+        double v_out[],
+        const double v0[], const double v1[], const uint dims)
+{
+	double d = 0.0;
+	for (uint j = 0; j < dims; j++) {
+		double a = v0[j] - v1[j];
+		d += sq(a);
+		v_out[j] = a;
+	}
+	if (d != 0.0 && ((d = sqrt(d)) != 0.0)) {
+		imul_vn_fl(v_out, 1.0 / d, dims);
+	}
+	return d;
+}
+
+MINLINE bool is_almost_zero_ex(double val, double eps)
+{
+	return (-eps < val) && (val < eps);
+}
+
+MINLINE bool is_almost_zero(double val)
+{
+	return is_almost_zero_ex(val, 1e-8);
+}
+
+MINLINE bool equals_vnvn(
+		const double v0[], const double v1[], const uint dims)
+{
+	for (uint j = 0; j < dims; j++) {
+		if (v0[j] != v1[j]) {
+			return false;
+		}
+	}
+	return true;
+}
+
+/** \} */

extern/libopenjpeg/bio.c

@@ -42,7 +42,7 @@ Write a bit
 @param bio BIO handle
 @param b Bit to write (0 or 1)
 */
-static void bio_putbit(opj_bio_t *bio, int b);
+static void bio_putbit(opj_bio_t *bio, unsigned int b);
 /**
 Read a bit
 @param bio BIO handle
@@ -78,7 +78,7 @@ static int bio_byteout(opj_bio_t *bio) {
 	if (bio->bp >= bio->end) {
 		return 1;
 	}
-	*bio->bp++ = bio->buf >> 8;
+	*bio->bp++ = (unsigned char)(bio->buf >> 8);
 	return 0;
 }
 
@@ -92,7 +92,7 @@ static int bio_bytein(opj_bio_t *bio) {
 	return 0;
 }
 
-static void bio_putbit(opj_bio_t *bio, int b) {
+static void bio_putbit(opj_bio_t *bio, unsigned int b) {
 	if (bio->ct == 0) {
 		bio_byteout(bio);
 	}
@@ -126,7 +126,7 @@ void bio_destroy(opj_bio_t *bio) {
 }
 
 int bio_numbytes(opj_bio_t *bio) {
-	return (bio->bp - bio->start);
+	return (int)(bio->bp - bio->start);
 }
 
 void bio_init_enc(opj_bio_t *bio, unsigned char *bp, int len) {

extern/libopenjpeg/cidx_manager.c

@@ -29,8 +29,6 @@
  * POSSIBILITY OF SUCH DAMAGE.
  */
 
-#include <stdio.h>
-#include <stdlib.h>
 #include "opj_includes.h"
 
 

extern/libopenjpeg/cio.c

@@ -30,6 +30,7 @@
  */
 
 #include "opj_includes.h"
+#include <assert.h>
 
 /* ----------------------------------------------------------------------- */
 
@@ -106,6 +107,7 @@ int OPJ_CALLCONV cio_tell(opj_cio_t *cio) {
  * pos : position, in number of bytes, from the beginning of the stream
  */
 void OPJ_CALLCONV cio_seek(opj_cio_t *cio, int pos) {
+  assert((cio->start + pos) <= cio->end);
 	cio->bp = cio->start + pos;
 }
 
@@ -113,6 +115,7 @@ void OPJ_CALLCONV cio_seek(opj_cio_t *cio, int pos) {
  * Number of bytes left before the end of the stream.
  */
 int cio_numbytesleft(opj_cio_t *cio) {
+  assert((cio->end - cio->bp) >= 0);
 	return cio->end - cio->bp;
 }
 
@@ -139,6 +142,7 @@ opj_bool cio_byteout(opj_cio_t *cio, unsigned char v) {
  * Read a byte.
  */
 unsigned char cio_bytein(opj_cio_t *cio) {
+  assert(cio->bp >= cio->start);
 	if (cio->bp >= cio->end) {
 		opj_event_msg(cio->cinfo, EVT_ERROR, "read error: passed the end of the codestream (start = %d, current = %d, end = %d\n", cio->start, cio->bp, cio->end);
 		return 0;
@@ -152,7 +156,7 @@ unsigned char cio_bytein(opj_cio_t *cio) {
  * v : value to write
  * n : number of bytes to write
  */
-unsigned int cio_write(opj_cio_t *cio, unsigned long long int v, int n) {
+unsigned int cio_write(opj_cio_t *cio, unsigned int64 v, int n) {
 	int i;
 	for (i = n - 1; i >= 0; i--) {
 		if( !cio_byteout(cio, (unsigned char) ((v >> (i << 3)) & 0xff)) )
@@ -173,7 +177,7 @@ unsigned int cio_read(opj_cio_t *cio, int n) {
 	unsigned int v;
 	v = 0;
 	for (i = n - 1; i >= 0; i--) {
-		v += cio_bytein(cio) << (i << 3);
+		v += (unsigned int)cio_bytein(cio) << (i << 3);
 	}
 	return v;
 }
@@ -184,6 +188,10 @@ unsigned int cio_read(opj_cio_t *cio, int n) {
  * n : number of bytes to skip
  */
 void cio_skip(opj_cio_t *cio, int n) {
+  assert((cio->bp + n) >= cio->bp);
+  if (((cio->bp + n) < cio->start) || ((cio->bp + n) > cio->end)) {
+    assert(0);
+  }
 	cio->bp += n;
 }
 

extern/libopenjpeg/cio.h

@@ -31,6 +31,13 @@
 
 #ifndef __CIO_H
 #define __CIO_H
+
+#if defined(_MSC_VER) || defined(__BORLANDC__)
+#define int64 __int64
+#else
+#define int64 long long
+#endif
+
 /**
 @file cio.h
 @brief Implementation of a byte input-output process (CIO)
@@ -63,7 +70,7 @@ Write some bytes
 @param n Number of bytes to write
 @return Returns the number of bytes written or 0 if an error occured
 */
-unsigned int cio_write(opj_cio_t *cio, unsigned long long int v, int n);
+unsigned int cio_write(opj_cio_t *cio, unsigned int64 v, int n);
 /**
 Read some bytes
 @param cio CIO handle

extern/libopenjpeg/event.c

@@ -103,18 +103,17 @@ opj_bool opj_event_msg(opj_common_ptr cinfo, int event_type, const char *fmt, ..
 		va_list arg;
 		int str_length/*, i, j*/; /* UniPG */
 		char message[MSG_SIZE];
-		memset(message, 0, MSG_SIZE);
 		/* initialize the optional parameter list */
 		va_start(arg, fmt);
-		/* check the length of the format string */
-		str_length = (strlen(fmt) > MSG_SIZE) ? MSG_SIZE : strlen(fmt);
 		/* parse the format string and put the result in 'message' */
-		vsprintf(message, fmt, arg); /* UniPG */
+		str_length = vsnprintf(message, MSG_SIZE, fmt, arg); /* UniPG */
 		/* deinitialize the optional parameter list */
 		va_end(arg);
 
 		/* output the message to the user program */
-		msg_handler(message, cinfo->client_data);
+    if( str_length > -1 && str_length < MSG_SIZE )
+      msg_handler(message, cinfo->client_data);
+    else return OPJ_FALSE;
 	}
 
 	return OPJ_TRUE;

extern/libopenjpeg/image.c

@@ -40,7 +40,7 @@ opj_image_t* OPJ_CALLCONV opj_image_create(int numcmpts, opj_image_cmptparm_t *c
 		image->color_space = clrspc;
 		image->numcomps = numcmpts;
 		/* allocate memory for the per-component information */
-		image->comps = (opj_image_comp_t*)opj_malloc(image->numcomps * sizeof(opj_image_comp_t));
+		image->comps = (opj_image_comp_t*)opj_calloc(1,image->numcomps * sizeof(opj_image_comp_t));
 		if(!image->comps) {
 			fprintf(stderr,"Unable to allocate memory for image.\n");
 			opj_image_destroy(image);
@@ -86,3 +86,4 @@ void OPJ_CALLCONV opj_image_destroy(opj_image_t *image) {
 		opj_free(image);
 	}
 }
+

extern/libopenjpeg/j2k.c

@@ -32,6 +32,7 @@
  */
 
 #include "opj_includes.h"
+#include <assert.h>
 
 /** @defgroup J2K J2K - JPEG-2000 codestream reader/writer */
 /*@{*/
@@ -404,6 +405,7 @@ static void j2k_write_siz(opj_j2k_t *j2k) {
 
 static void j2k_read_siz(opj_j2k_t *j2k) {
 	int len, i;
+  int n_comps;
 	
 	opj_cio_t *cio = j2k->cio;
 	opj_image_t *image = j2k->image;
@@ -422,12 +424,33 @@ static void j2k_read_siz(opj_j2k_t *j2k) {
 	
 	if ((image->x0<0)||(image->x1<0)||(image->y0<0)||(image->y1<0)) {
 		opj_event_msg(j2k->cinfo, EVT_ERROR,
-									"%s: invalid image size (x0:%d, x1:%d, y0:%d, y1:%d)\n",
+									"invalid image size (x0:%d, x1:%d, y0:%d, y1:%d)\n",
 									image->x0,image->x1,image->y0,image->y1);
 		return;
 	}
 	
+  n_comps = (len - 36 - 2 ) / 3;
+  assert( (len - 36 - 2 ) % 3 == 0 );
 	image->numcomps = cio_read(cio, 2);	/* Csiz */
+  assert( n_comps == image->numcomps );
+  (void)n_comps;
+
+  /* testcase 4035.pdf.SIGSEGV.d8b.3375 */
+  if (image->x0 > image->x1 || image->y0 > image->y1) {
+    opj_event_msg(j2k->cinfo, EVT_ERROR, "Error with SIZ marker: negative image size (%d x %d)\n", image->x1 - image->x0, image->y1 - image->y0);
+    return;
+  }
+  /* testcase 2539.pdf.SIGFPE.706.1712 (also 3622.pdf.SIGFPE.706.2916 and 4008.pdf.SIGFPE.706.3345 and maybe more) */
+  if (!(cp->tdx * cp->tdy)) {
+    opj_event_msg(j2k->cinfo, EVT_ERROR, "Error with SIZ marker: invalid tile size (tdx: %d, tdy: %d)\n", cp->tdx, cp->tdy);
+    return;
+  }
+
+  /* testcase 1610.pdf.SIGSEGV.59c.681 */
+  if (((int64)image->x1) * ((int64)image->y1) != (image->x1 * image->y1)) {
+    opj_event_msg(j2k->cinfo, EVT_ERROR, "Prevent buffer overflow (x1: %d, y1: %d)\n", image->x1, image->y1);
+    return;
+  }
 
 #ifdef USE_JPWL
 	if (j2k->cp->correct) {
@@ -466,11 +489,19 @@ static void j2k_read_siz(opj_j2k_t *j2k) {
 		/* update components number in the jpwl_exp_comps filed */
 		cp->exp_comps = image->numcomps;
 	}
+#else
+  (void)len;
 #endif /* USE_JPWL */
 
+  /* prevent division by zero */
+  if (!(cp->tdx * cp->tdy)) {
+    opj_event_msg(j2k->cinfo, EVT_ERROR, "invalid tile size (tdx: %d, tdy: %d)\n", cp->tdx, cp->tdy);
+    return;
+  }
+
 	image->comps = (opj_image_comp_t*) opj_calloc(image->numcomps, sizeof(opj_image_comp_t));
 	for (i = 0; i < image->numcomps; i++) {
-		int tmp, w, h;
+		int tmp;
 		tmp = cio_read(cio, 1);		/* Ssiz_i */
 		image->comps[i].prec = (tmp & 0x7f) + 1;
 		image->comps[i].sgnd = tmp >> 7;
@@ -506,9 +537,11 @@ static void j2k_read_siz(opj_j2k_t *j2k) {
 		}
 #endif /* USE_JPWL */
 
-		/* TODO: unused ? */
-		w = int_ceildiv(image->x1 - image->x0, image->comps[i].dx);
-		h = int_ceildiv(image->y1 - image->y0, image->comps[i].dy);
+    /* prevent division by zero */
+    if (!(image->comps[i].dx * image->comps[i].dy)) {
+      opj_event_msg(j2k->cinfo, EVT_ERROR, "JPWL: invalid component size (dx: %d, dy: %d)\n", image->comps[i].dx, image->comps[i].dy);
+      return;
+    }
 
 		image->comps[i].resno_decoded = 0;	/* number of resolution decoded */
 		image->comps[i].factor = cp->reduce; /* reducing factor per component */
@@ -517,6 +550,15 @@ static void j2k_read_siz(opj_j2k_t *j2k) {
 	cp->tw = int_ceildiv(image->x1 - cp->tx0, cp->tdx);
 	cp->th = int_ceildiv(image->y1 - cp->ty0, cp->tdy);
 
+  /* gdal_fuzzer_check_number_of_tiles.jp2 */
+  if (cp->tw == 0 || cp->th == 0 || cp->tw > 65535 / cp->th) {
+    opj_event_msg(j2k->cinfo, EVT_ERROR, 
+                            "Invalid number of tiles : %u x %u (maximum fixed by jpeg2000 norm is 65535 tiles)\n",
+                            cp->tw, cp->th);
+    return;
+  }
+
+
 #ifdef USE_JPWL
 	if (j2k->cp->correct) {
 		/* if JPWL is on, we check whether TX errors have damaged
@@ -558,7 +600,17 @@ static void j2k_read_siz(opj_j2k_t *j2k) {
 #endif /* USE_JPWL */
 
 	cp->tcps = (opj_tcp_t*) opj_calloc(cp->tw * cp->th, sizeof(opj_tcp_t));
+    if (cp->tcps == NULL)
+    {
+        opj_event_msg(j2k->cinfo, EVT_ERROR, "Out of memory\n");
+        return;
+    }
 	cp->tileno = (int*) opj_malloc(cp->tw * cp->th * sizeof(int));
+    if (cp->tileno == NULL)
+    {
+        opj_event_msg(j2k->cinfo, EVT_ERROR, "Out of memory\n");
+        return;
+    }
 	cp->tileno_size = 0;
 	
 #ifdef USE_JPWL
@@ -684,6 +736,12 @@ static void j2k_read_cox(opj_j2k_t *j2k, int compno) {
 					"of resolutions of this component\nModify the cp_reduce parameter.\n\n", compno);
 		j2k->state |= J2K_STATE_ERR;
 	}
+  if( tccp->numresolutions > J2K_MAXRLVLS ) {
+    opj_event_msg(j2k->cinfo, EVT_ERROR, "Error decoding component %d.\nThe number of resolutions is too big: %d vs max= %d. Truncating.\n\n",
+      compno, tccp->numresolutions, J2K_MAXRLVLS);
+		j2k->state |= J2K_STATE_ERR;
+    tccp->numresolutions = J2K_MAXRLVLS;
+ }
 
 	tccp->cblkw = cio_read(cio, 1) + 2;	/* SPcox (E) */
 	tccp->cblkh = cio_read(cio, 1) + 2;	/* SPcox (F) */
@@ -753,6 +811,7 @@ static void j2k_read_cod(opj_j2k_t *j2k) {
 	opj_image_t *image = j2k->image;
 	
 	len = cio_read(cio, 2);				/* Lcod */
+  (void)len;
 	tcp->csty = cio_read(cio, 1);		/* Scod */
 	tcp->prg = (OPJ_PROG_ORDER)cio_read(cio, 1);		/* SGcod (A) */
 	tcp->numlayers = cio_read(cio, 2);	/* SGcod (B) */
@@ -806,7 +865,14 @@ static void j2k_read_coc(opj_j2k_t *j2k) {
 	opj_cio_t *cio = j2k->cio;
 	
 	len = cio_read(cio, 2);		/* Lcoc */
+  (void)len;
 	compno = cio_read(cio, image->numcomps <= 256 ? 1 : 2);	/* Ccoc */
+  if (compno >= image->numcomps) {
+    opj_event_msg(j2k->cinfo, EVT_ERROR,
+      "bad component number in COC (%d out of a maximum of %d)\n",
+      compno, image->numcomps);
+    return;
+  }
 	tcp->tccps[compno].csty = cio_read(cio, 1);	/* Scoc */
 	j2k_read_cox(j2k, compno);
 }
@@ -877,6 +943,8 @@ static void j2k_read_qcx(opj_j2k_t *j2k, int compno, int len) {
 		opj_event_msg(j2k->cinfo, EVT_WARNING ,
 					"bad number of subbands in Sqcx (%d) regarding to J2K_MAXBANDS (%d) \n"
 				    "- limiting number of bands to J2K_MAXBANDS and try to move to the next markers\n", numbands, J2K_MAXBANDS);
+    /* edf_c2_1013627.jp2 */
+    numbands = 1;
 	}
 
 #endif /* USE_JPWL */
@@ -988,9 +1056,16 @@ static void j2k_read_qcc(opj_j2k_t *j2k) {
 
 		/* keep your private count of tiles */
 		backup_compno++;
-	};
+	}
 #endif /* USE_JPWL */
 
+  if ((compno < 0) || (compno >= numcomp)) {
+    opj_event_msg(j2k->cinfo, EVT_ERROR,
+      "bad component number in QCC (%d out of a maximum of %d)\n",
+      compno, j2k->image->numcomps);
+    return;
+  }
+
 	j2k_read_qcx(j2k, compno, len - 2 - (numcomp <= 256 ? 1 : 2));
 }
 
@@ -1036,6 +1111,15 @@ static void j2k_read_poc(opj_j2k_t *j2k) {
 	len = cio_read(cio, 2);		/* Lpoc */
 	numpchgs = (len - 2) / (5 + 2 * (numcomps <= 256 ? 1 : 2));
 	
+  if( numpchgs >= 32 )
+    {
+    /* edf_c2_1103421.jp2 */
+    opj_event_msg(j2k->cinfo, EVT_ERROR,
+      "bad number of POCS (%d out of a maximum of %d)\n",
+      numpchgs, 32);
+    numpchgs = 0;
+    }
+
 	for (i = old_poc; i < numpchgs + old_poc; i++) {
 		opj_poc_t *poc;
 		poc = &tcp->pocs[i];
@@ -1058,9 +1142,12 @@ static void j2k_read_crg(opj_j2k_t *j2k) {
 	int numcomps = j2k->image->numcomps;
 	
 	len = cio_read(cio, 2);			/* Lcrg */
+  (void)len;
 	for (i = 0; i < numcomps; i++) {
 		Xcrg_i = cio_read(cio, 2);	/* Xcrg_i */
+    (void)Xcrg_i;
 		Ycrg_i = cio_read(cio, 2);	/* Ycrg_i */
+    (void)Ycrg_i;
 	}
 }
 
@@ -1072,13 +1159,16 @@ static void j2k_read_tlm(opj_j2k_t *j2k) {
 	
 	len = cio_read(cio, 2);		/* Ltlm */
 	Ztlm = cio_read(cio, 1);	/* Ztlm */
+  (void)Ztlm;
 	Stlm = cio_read(cio, 1);	/* Stlm */
 	ST = ((Stlm >> 4) & 0x01) + ((Stlm >> 4) & 0x02);
 	SP = (Stlm >> 6) & 0x01;
 	tile_tlm = (len - 4) / ((SP + 1) * 2 + ST);
 	for (i = 0; i < tile_tlm; i++) {
 		Ttlm_i = cio_read(cio, ST);	/* Ttlm_i */
+    (void)Ttlm_i;
 		Ptlm_i = cio_read(cio, SP ? 4 : 2);	/* Ptlm_i */
+    (void)Ptlm_i;
 	}
 }
 
@@ -1089,6 +1179,7 @@ static void j2k_read_plm(opj_j2k_t *j2k) {
 
 	len = cio_read(cio, 2);		/* Lplm */
 	Zplm = cio_read(cio, 1);	/* Zplm */
+  (void)Zplm;
 	len -= 3;
 	while (len > 0) {
 		Nplm = cio_read(cio, 4);		/* Nplm */
@@ -1114,6 +1205,7 @@ static void j2k_read_plt(opj_j2k_t *j2k) {
 	
 	len = cio_read(cio, 2);		/* Lplt */
 	Zplt = cio_read(cio, 1);	/* Zplt */
+  (void)Zplt;
 	for (i = len - 3; i > 0; i--) {
 		add = cio_read(cio, 1);
 		packet_len = (packet_len << 7) + add;	/* Iplt_i */
@@ -1261,6 +1353,7 @@ static void j2k_read_sot(opj_j2k_t *j2k) {
 	opj_cio_t *cio = j2k->cio;
 
 	len = cio_read(cio, 2);
+  (void)len;
 	tileno = cio_read(cio, 2);
 
 #ifdef USE_JPWL
@@ -1269,7 +1362,7 @@ static void j2k_read_sot(opj_j2k_t *j2k) {
 		static int backup_tileno = 0;
 
 		/* tileno is negative or larger than the number of tiles!!! */
-		if ((tileno < 0) || (tileno > (cp->tw * cp->th))) {
+		if ((tileno < 0) || (tileno >= (cp->tw * cp->th))) {
 			opj_event_msg(j2k->cinfo, EVT_ERROR,
 				"JPWL: bad tile number (%d out of a maximum of %d)\n",
 				tileno, (cp->tw * cp->th));
@@ -1286,8 +1379,18 @@ static void j2k_read_sot(opj_j2k_t *j2k) {
 
 		/* keep your private count of tiles */
 		backup_tileno++;
-	};
+	}
+  else
 #endif /* USE_JPWL */
+  {
+    /* tileno is negative or larger than the number of tiles!!! */
+    if ((tileno < 0) || (tileno >= (cp->tw * cp->th))) {
+      opj_event_msg(j2k->cinfo, EVT_ERROR,
+        "JPWL: bad tile number (%d out of a maximum of %d)\n",
+        tileno, (cp->tw * cp->th));
+      return;
+    }
+  }
 	
 	if (cp->tileno_size == 0) {
 		cp->tileno[cp->tileno_size] = tileno;
@@ -1325,8 +1428,18 @@ static void j2k_read_sot(opj_j2k_t *j2k) {
 				totlen);
 		}
 
-	};
+	}
+  else
 #endif /* USE_JPWL */
+  {
+    /* totlen is negative or larger than the bytes left!!! */
+    if ((totlen < 0) || (totlen > (cio_numbytesleft(cio) + 8))) {
+      opj_event_msg(j2k->cinfo, EVT_ERROR,
+        "JPWL: bad tile byte size (%d bytes against %d bytes left)\n",
+        totlen, cio_numbytesleft(cio) + 8);
+      return;
+    }
+  }
 
 	if (!totlen)
 		totlen = cio_numbytesleft(cio) + 8;
@@ -1478,7 +1591,13 @@ static void j2k_read_sod(opj_j2k_t *j2k) {
 	}	
 
 	data = j2k->tile_data[curtileno];
+  data_ptr = data; /* store in case of failure */
 	data = (unsigned char*) opj_realloc(data, (j2k->tile_len[curtileno] + len) * sizeof(unsigned char));
+  if( data == NULL ) {
+    opj_event_msg(j2k->cinfo, EVT_ERROR, "Could not reallocated\n" );
+    opj_free( data_ptr );
+    return;
+    }
 
 	data_ptr = data + j2k->tile_len[curtileno];
 	for (i = 0; i < len; i++) {
@@ -1518,8 +1637,10 @@ static void j2k_read_rgn(opj_j2k_t *j2k) {
 	int numcomps = j2k->image->numcomps;
 
 	len = cio_read(cio, 2);										/* Lrgn */
+  (void)len;
 	compno = cio_read(cio, numcomps <= 256 ? 1 : 2);			/* Crgn */
 	roisty = cio_read(cio, 1);									/* Srgn */
+  (void)roisty;
 
 #ifdef USE_JPWL
 	if (j2k->cp->correct) {
@@ -1536,6 +1657,13 @@ static void j2k_read_rgn(opj_j2k_t *j2k) {
 	};
 #endif /* USE_JPWL */
 
+  if (compno >= numcomps) {
+    opj_event_msg(j2k->cinfo, EVT_ERROR,
+      "bad component number in RGN (%d out of a maximum of %d)\n",
+      compno, j2k->image->numcomps);
+    return;
+  }
+
 	tcp->tccps[compno].roishift = cio_read(cio, 1);				/* SPrgn */
 }
 
@@ -1554,7 +1682,7 @@ static void j2k_write_eoc(opj_j2k_t *j2k) {
 
 static void j2k_read_eoc(opj_j2k_t *j2k) {
 	int i, tileno;
-	opj_bool success;
+	opj_bool success = OPJ_FALSE;
 
 	/* if packets should be decoded */
 	if (j2k->cp->limit_decoding != DECODE_ALL_BUT_PACKETS) {
@@ -1562,11 +1690,17 @@ static void j2k_read_eoc(opj_j2k_t *j2k) {
 		tcd_malloc_decode(tcd, j2k->image, j2k->cp);
 		for (i = 0; i < j2k->cp->tileno_size; i++) {
 			tcd_malloc_decode_tile(tcd, j2k->image, j2k->cp, i, j2k->cstr_info);
-			tileno = j2k->cp->tileno[i];
-			success = tcd_decode_tile(tcd, j2k->tile_data[tileno], j2k->tile_len[tileno], tileno, j2k->cstr_info);
-			opj_free(j2k->tile_data[tileno]);
-			j2k->tile_data[tileno] = NULL;
-			tcd_free_decode_tile(tcd, i);
+			if (j2k->cp->tileno[i] != -1)
+			{
+				tileno = j2k->cp->tileno[i];
+				success = tcd_decode_tile(tcd, j2k->tile_data[tileno], j2k->tile_len[tileno], tileno, j2k->cstr_info);
+        assert( tileno != -1 );
+				opj_free(j2k->tile_data[tileno]);
+				j2k->tile_data[tileno] = NULL;
+				tcd_free_decode_tile(tcd, i);
+			}
+			else
+				success = OPJ_FALSE;
 			if (success == OPJ_FALSE) {
 				j2k->state |= J2K_STATE_ERR;
 				break;
@@ -1737,6 +1871,17 @@ void j2k_destroy_decompress(opj_j2k_t *j2k) {
 		opj_free(j2k->tile_len);
 	}
 	if(j2k->tile_data != NULL) {
+        if(j2k->cp != NULL) {
+            for (i = 0; i < j2k->cp->tileno_size; i++) {
+                int tileno = j2k->cp->tileno[i];
+                if( tileno != -1 )
+                  {
+                  opj_free(j2k->tile_data[tileno]);
+                  j2k->tile_data[tileno] = NULL;
+                  }
+            }
+        }
+
 		opj_free(j2k->tile_data);
 	}
 	if(j2k->default_tcp != NULL) {
@@ -1856,9 +2001,15 @@ opj_image_t* j2k_decode(opj_j2k_t *j2k, opj_cio_t *cio, opj_codestream_info_t *c
 #endif /* USE_JPWL */
 
 		if (id >> 8 != 0xff) {
-			opj_image_destroy(image);
-			opj_event_msg(cinfo, EVT_ERROR, "%.8x: expected a marker instead of %x\n", cio_tell(cio) - 2, id);
-			return 0;
+		if(cio_numbytesleft(cio) != 0) /* not end of file reached and no EOC */
+	   {
+		opj_event_msg(cinfo, EVT_ERROR, "%.8x: expected a marker instead of %x\n", cio_tell(cio) - 2, id);
+		opj_image_destroy(image);
+		return 0;
+	   }
+		opj_event_msg(cinfo, EVT_WARNING, "%.8x: expected a marker instead of %x\n", cio_tell(cio) - 2, id);
+		j2k->state = J2K_STATE_NEOC;
+		break;
 		}
 		e = j2k_dec_mstab_lookup(id);
 		/* Check if the marker is known*/
@@ -1877,7 +2028,10 @@ opj_image_t* j2k_decode(opj_j2k_t *j2k, opj_cio_t *cio, opj_codestream_info_t *c
 			(*e->handler)(j2k);
 		}
 		if (j2k->state & J2K_STATE_ERR) 
+        {
+            opj_image_destroy(image);
 			return NULL;	
+        }
 
 		if (j2k->state == J2K_STATE_MT) {
 			break;
@@ -1888,6 +2042,11 @@ opj_image_t* j2k_decode(opj_j2k_t *j2k, opj_cio_t *cio, opj_codestream_info_t *c
 	}
 	if (j2k->state == J2K_STATE_NEOC) {
 		j2k_read_eoc(j2k);
+		/* Check one last time for errors during decoding before returning */
+		if (j2k->state & J2K_STATE_ERR) {
+			opj_image_destroy(image);
+			return NULL;
+		}
 	}
 
 	if (j2k->state != J2K_STATE_MT) {
@@ -1949,9 +2108,15 @@ opj_image_t* j2k_decode_jpt_stream(opj_j2k_t *j2k, opj_cio_t *cio,  opj_codestre
 		
 		id = cio_read(cio, 2);
 		if (id >> 8 != 0xff) {
-			opj_image_destroy(image);
-			opj_event_msg(cinfo, EVT_ERROR, "%.8x: expected a marker instead of %x\n", cio_tell(cio) - 2, id);
-			return 0;
+        if(cio_numbytesleft(cio) != 0) /* no end of file reached and no EOC */
+	  {
+		opj_event_msg(cinfo, EVT_ERROR, "%.8x: expected a marker instead of %x\n", cio_tell(cio) - 2, id);
+		opj_image_destroy(image);
+		return 0;
+	  }
+		opj_event_msg(cinfo, EVT_WARNING, "%.8x: expected a marker instead of %x\n", cio_tell(cio) - 2, id);
+		j2k->state = J2K_STATE_NEOC;
+		break;
 		}
 		e = j2k_dec_mstab_lookup(id);
 		if (!(j2k->state & e->states)) {

extern/libopenjpeg/jp2.c

@@ -30,6 +30,7 @@
  * POSSIBILITY OF SUCH DAMAGE.
  */
 #include "opj_includes.h"
+#include <assert.h>
 
 /** @defgroup JP2 JP2 - JPEG-2000 file format reader/writer */
 /*@{*/
@@ -172,6 +173,9 @@ static opj_bool jp2_read_boxhdr(opj_common_ptr cinfo, opj_cio_t *cio, opj_jp2_bo
 	}
 	else if (box->length == 0) {
 		box->length = cio_numbytesleft(cio) + 8;
+	} else if (box->length < 0) {
+		opj_event_msg(cinfo, EVT_ERROR, "Integer overflow in box->length\n");
+		return OPJ_FALSE; /* TODO: actually check jp2_read_boxhdr's return value */
 	}
 	
 	return OPJ_TRUE;
@@ -206,7 +210,10 @@ static opj_bool jp2_read_ihdr(opj_jp2_t *jp2, opj_cio_t *cio) {
 
 	opj_common_ptr cinfo = jp2->cinfo;
 
-	jp2_read_boxhdr(cinfo, cio, &box);
+  if(jp2_read_boxhdr(cinfo, cio, &box) == OPJ_FALSE ) {
+    opj_event_msg(cinfo, EVT_ERROR, "Failed to read boxhdr\n");
+    return OPJ_FALSE;
+  }
 	if (JP2_IHDR != box.type) {
 		opj_event_msg(cinfo, EVT_ERROR, "Expected IHDR Marker\n");
 		return OPJ_FALSE;
@@ -279,7 +286,10 @@ static opj_bool jp2_read_bpcc(opj_jp2_t *jp2, opj_cio_t *cio) {
 
 	opj_common_ptr cinfo = jp2->cinfo;
 
-	jp2_read_boxhdr(cinfo, cio, &box);
+  if( jp2_read_boxhdr(cinfo, cio, &box) == OPJ_FALSE ) {
+    opj_event_msg(cinfo, EVT_ERROR, "Failed to read boxhdr\n");
+    return OPJ_FALSE;
+  }
 	if (JP2_BPCC != box.type) {
 		opj_event_msg(cinfo, EVT_ERROR, "Expected BPCC Marker\n");
 		return OPJ_FALSE;
@@ -469,7 +479,7 @@ static opj_bool jp2_read_pclr(opj_jp2_t *jp2, opj_cio_t *cio,
 	for(i = 0; i < nr_channels; ++i)
   {
 /* Cji */
-	*entries++ = cio_read(cio, channel_size[i]>>3);
+	*entries++ = cio_read(cio, (channel_size[i]+7)>>3);
   }
    }
 
@@ -512,10 +522,8 @@ static opj_bool jp2_read_cmap(opj_jp2_t *jp2, opj_cio_t *cio,
 static void jp2_apply_cdef(opj_image_t *image, opj_jp2_color_t *color)
 {
 	opj_jp2_cdef_info_t *info;
-	int color_space;
 	unsigned short i, n, cn, typ, asoc, acn;
 
-	color_space = image->color_space;
 	info = color->jp2_cdef->info;
 	n = color->jp2_cdef->n;
 
@@ -525,6 +533,7 @@ static void jp2_apply_cdef(opj_image_t *image, opj_jp2_color_t *color)
 	if((asoc = info[i].asoc) == 0) continue;
 
 	cn = info[i].cn; typ = info[i].typ; acn = asoc - 1;
+  (void)typ;
 
 	if(cn != acn)
   {
@@ -639,90 +648,173 @@ opj_bool jp2_read_jp2h(opj_jp2_t *jp2, opj_cio_t *cio, opj_jp2_color_t *color)
 
 	opj_common_ptr cinfo = jp2->cinfo;
 
-	jp2_read_boxhdr(cinfo, cio, &box);
-	do 
-   {
-	if (JP2_JP2H != box.type) 
-  {
-	if (box.type == JP2_JP2C) 
- {
-	opj_event_msg(cinfo, EVT_ERROR, "Expected JP2H Marker\n");
-	return OPJ_FALSE;
- }
-	cio_skip(cio, box.length - 8);
+  if( jp2_read_boxhdr(cinfo, cio, &box) == OPJ_FALSE ) return OPJ_FALSE;
+  do {
+    if (JP2_JP2H != box.type) 
+      {
+      if (box.type == JP2_JP2C) 
+        {
+        opj_event_msg(cinfo, EVT_ERROR, "Expected JP2H Marker\n");
+        return OPJ_FALSE;
+        }
+	  if (box.length <= 8) return OPJ_FALSE;
+      cio_skip(cio, box.length - 8);
 
-	if(cio->bp >= cio->end) return OPJ_FALSE;
+      if(cio->bp >= cio->end) return OPJ_FALSE;
 
-	jp2_read_boxhdr(cinfo, cio, &box);
-  }
-   } while(JP2_JP2H != box.type);
+      if( jp2_read_boxhdr(cinfo, cio, &box) == OPJ_FALSE ) return OPJ_FALSE;
+      }
+  } while(JP2_JP2H != box.type);
 
 	if (!jp2_read_ihdr(jp2, cio))
 		return OPJ_FALSE;
 	jp2h_end = box.init_pos + box.length;
 
-	if (jp2->bpc == 255) 
-   {
-	if (!jp2_read_bpcc(jp2, cio))
-		return OPJ_FALSE;
-   }
-	jp2_read_boxhdr(cinfo, cio, &box);
+  if (jp2->bpc == 255) 
+    {
+    if (!jp2_read_bpcc(jp2, cio))
+      return OPJ_FALSE;
+    }
+  if( jp2_read_boxhdr(cinfo, cio, &box) == OPJ_FALSE ) return OPJ_FALSE;
 
-	while(cio_tell(cio) < jp2h_end)
-   {
-	if(box.type == JP2_COLR)
-  {
-	if( !jp2_read_colr(jp2, cio, &box, color))
- {
-    cio_seek(cio, box.init_pos + 8);
-    cio_skip(cio, box.length - 8);
- }
-    jp2_read_boxhdr(cinfo, cio, &box);
-    continue;
-  }
+  while(cio_tell(cio) < jp2h_end)
+    {
+    if(box.type == JP2_COLR)
+      {
+      if( !jp2_read_colr(jp2, cio, &box, color))
+        {
+        if (box.length <= 8) return OPJ_FALSE;
+        cio_seek(cio, box.init_pos + 8);
+        cio_skip(cio, box.length - 8);
+        }
+      if( jp2_read_boxhdr(cinfo, cio, &box) == OPJ_FALSE ) return OPJ_FALSE;
+      continue;
+      }
     if(box.type == JP2_CDEF && !jp2->ignore_pclr_cmap_cdef)
-  {
-    if( !jp2_read_cdef(jp2, cio, &box, color))
- {
-    cio_seek(cio, box.init_pos + 8);
-    cio_skip(cio, box.length - 8);
- }
-    jp2_read_boxhdr(cinfo, cio, &box);
-    continue;
-  }
+      {
+      if( !jp2_read_cdef(jp2, cio, &box, color))
+        {
+        if (box.length <= 8) return OPJ_FALSE;
+        cio_seek(cio, box.init_pos + 8);
+        cio_skip(cio, box.length - 8);
+        }
+      if( jp2_read_boxhdr(cinfo, cio, &box) == OPJ_FALSE ) return OPJ_FALSE;
+      continue;
+      }
     if(box.type == JP2_PCLR && !jp2->ignore_pclr_cmap_cdef)
-  {
-    if( !jp2_read_pclr(jp2, cio, &box, color))
- {
-    cio_seek(cio, box.init_pos + 8);
-    cio_skip(cio, box.length - 8);
- }
-    jp2_read_boxhdr(cinfo, cio, &box);
-    continue;
-  }
+      {
+      if( !jp2_read_pclr(jp2, cio, &box, color))
+        {
+        if (box.length <= 8) return OPJ_FALSE;
+        cio_seek(cio, box.init_pos + 8);
+        cio_skip(cio, box.length - 8);
+        }
+      if( jp2_read_boxhdr(cinfo, cio, &box) == OPJ_FALSE ) return OPJ_FALSE;
+      continue;
+      }
     if(box.type == JP2_CMAP && !jp2->ignore_pclr_cmap_cdef)
-  {
-    if( !jp2_read_cmap(jp2, cio, &box, color))
- {
+      {
+      if( !jp2_read_cmap(jp2, cio, &box, color))
+        {
+        if (box.length <= 8) return OPJ_FALSE;
+        cio_seek(cio, box.init_pos + 8);
+        cio_skip(cio, box.length - 8);
+        }
+      if( jp2_read_boxhdr(cinfo, cio, &box) == OPJ_FALSE ) return OPJ_FALSE;
+      continue;
+      }
+    if (box.length <= 8) return OPJ_FALSE;
     cio_seek(cio, box.init_pos + 8);
     cio_skip(cio, box.length - 8);
- }
-    jp2_read_boxhdr(cinfo, cio, &box);
-    continue;
-  }
-	cio_seek(cio, box.init_pos + 8);
-	cio_skip(cio, box.length - 8);
-	jp2_read_boxhdr(cinfo, cio, &box);
+    if( jp2_read_boxhdr(cinfo, cio, &box) == OPJ_FALSE ) return OPJ_FALSE;
 
-   }/* while(cio_tell(cio) < box_end) */
+    }/* while(cio_tell(cio) < box_end) */
 
-	cio_seek(cio, jp2h_end);
+  cio_seek(cio, jp2h_end);
 
-/* Part 1, I.5.3.3 : 'must contain at least one' */
-	return (color->jp2_has_colr == 1);
+  /* Part 1, I.5.3.3 : 'must contain at least one' */
+  return (color->jp2_has_colr == 1);
 
 }/* jp2_read_jp2h() */
 
+static opj_bool opj_jp2_check_color(opj_image_t *image, opj_jp2_color_t *color, opj_common_ptr cinfo)
+{
+	int i;
+
+	/* testcase 4149.pdf.SIGSEGV.cf7.3501 */
+	if (color->jp2_cdef) {
+		opj_jp2_cdef_info_t *info = color->jp2_cdef->info;
+		int n = color->jp2_cdef->n;
+
+		for (i = 0; i < n; i++) {
+			if (info[i].cn >= image->numcomps) {
+				opj_event_msg(cinfo, EVT_ERROR, "Invalid component index %d (>= %d).\n", info[i].cn, image->numcomps);
+				return OPJ_FALSE;
+			}
+			if (info[i].asoc > 0 && (info[i].asoc - 1) >= image->numcomps) {
+				opj_event_msg(cinfo, EVT_ERROR, "Invalid component index %d (>= %d).\n", info[i].asoc - 1, image->numcomps);
+				return OPJ_FALSE;
+			}
+		}
+	}
+
+	/* testcases 451.pdf.SIGSEGV.f4c.3723, 451.pdf.SIGSEGV.5b5.3723 and
+	   66ea31acbb0f23a2bbc91f64d69a03f5_signal_sigsegv_13937c0_7030_5725.pdf */
+	if (color->jp2_pclr && color->jp2_pclr->cmap) {
+		int nr_channels = color->jp2_pclr->nr_channels;
+		opj_jp2_cmap_comp_t *cmap = color->jp2_pclr->cmap;
+		opj_bool *pcol_usage, is_sane = OPJ_TRUE;
+
+		/* verify that all original components match an existing one */
+		for (i = 0; i < nr_channels; i++) {
+			if (cmap[i].cmp >= image->numcomps) {
+				opj_event_msg(cinfo, EVT_ERROR, "Invalid component index %d (>= %d).\n", cmap[i].cmp, image->numcomps);
+				is_sane = OPJ_FALSE;
+			}
+		}
+
+		pcol_usage = opj_calloc(nr_channels, sizeof(opj_bool));
+		if (!pcol_usage) {
+			opj_event_msg(cinfo, EVT_ERROR, "Unexpected OOM.\n");
+			return OPJ_FALSE;
+		}
+		/* verify that no component is targeted more than once */
+		for (i = 0; i < nr_channels; i++) {
+      int pcol = cmap[i].pcol;
+      assert(cmap[i].mtyp == 0 || cmap[i].mtyp == 1);
+			if (pcol >= nr_channels) {
+				opj_event_msg(cinfo, EVT_ERROR, "Invalid component/palette index for direct mapping %d.\n", pcol);
+				is_sane = OPJ_FALSE;
+			}
+			else if (pcol_usage[pcol] && cmap[i].mtyp == 1) {
+				opj_event_msg(cinfo, EVT_ERROR, "Component %d is mapped twice.\n", pcol);
+				is_sane = OPJ_FALSE;
+			}
+      else if (cmap[i].mtyp == 0 && cmap[i].pcol != 0) {
+        /* I.5.3.5 PCOL: If the value of the MTYP field for this channel is 0, then
+         * the value of this field shall be 0. */
+				opj_event_msg(cinfo, EVT_ERROR, "Direct use at #%d however pcol=%d.\n", i, pcol);
+				is_sane = OPJ_FALSE;
+      }
+			else
+				pcol_usage[pcol] = OPJ_TRUE;
+		}
+		/* verify that all components are targeted at least once */
+		for (i = 0; i < nr_channels; i++) {
+			if (!pcol_usage[i] && cmap[i].mtyp != 0) {
+				opj_event_msg(cinfo, EVT_ERROR, "Component %d doesn't have a mapping.\n", i);
+				is_sane = OPJ_FALSE;
+			}
+		}
+		opj_free(pcol_usage);
+		if (!is_sane) {
+			return OPJ_FALSE;
+		}
+	}
+
+	return OPJ_TRUE;
+}
+
 opj_image_t* opj_jp2_decode(opj_jp2_t *jp2, opj_cio_t *cio, 
 	opj_codestream_info_t *cstr_info) 
 {
@@ -756,6 +848,10 @@ opj_image_t* opj_jp2_decode(opj_jp2_t *jp2, opj_cio_t *cio,
    }
    
     if (!jp2->ignore_pclr_cmap_cdef){
+	    if (!opj_jp2_check_color(image, &color, cinfo)) {
+        opj_event_msg(cinfo, EVT_ERROR, "Failed to decode PCRL box\n");
+		    return NULL;
+	    }
 
     /* Set Image Color Space */
 	if (jp2->enumcs == 16)
@@ -839,8 +935,10 @@ static opj_bool jp2_read_ftyp(opj_jp2_t *jp2, opj_cio_t *cio) {
 
 	opj_common_ptr cinfo = jp2->cinfo;
 
-	jp2_read_boxhdr(cinfo, cio, &box);
-
+  if( jp2_read_boxhdr(cinfo, cio, &box) == OPJ_FALSE ) {
+    opj_event_msg(cinfo, EVT_ERROR, "Failed to read boxhdr\n");
+    return OPJ_FALSE;
+  }
 	if (JP2_FTYP != box.type) {
 		opj_event_msg(cinfo, EVT_ERROR, "Expected FTYP Marker\n");
 		return OPJ_FALSE;
@@ -849,6 +947,14 @@ static opj_bool jp2_read_ftyp(opj_jp2_t *jp2, opj_cio_t *cio) {
 	jp2->brand = cio_read(cio, 4);		/* BR */
 	jp2->minversion = cio_read(cio, 4);	/* MinV */
 	jp2->numcl = (box.length - 16) / 4;
+
+  /* edf_c2_1673169.jp2 */
+	if (cio_numbytesleft(cio) < ((int)jp2->numcl * 4)) {
+		opj_event_msg(cinfo, EVT_ERROR, "Not enough bytes in FTYP Box "
+				"(expected %d, but only %d left)\n",
+				((int)jp2->numcl * 4), cio_numbytesleft(cio));
+		return OPJ_FALSE;
+	}
 	jp2->cl = (unsigned int *) opj_malloc(jp2->numcl * sizeof(unsigned int));
 
 	for (i = 0; i < (int)jp2->numcl; i++) {
@@ -897,15 +1003,20 @@ static opj_bool jp2_read_jp2c(opj_jp2_t *jp2, opj_cio_t *cio, unsigned int *j2k_
 
 	opj_common_ptr cinfo = jp2->cinfo;
 
-	jp2_read_boxhdr(cinfo, cio, &box);
+  if( jp2_read_boxhdr(cinfo, cio, &box) == OPJ_FALSE ) {
+    opj_event_msg(cinfo, EVT_ERROR, "Failed to read boxhdr\n");
+    return OPJ_FALSE;
+  }
 	do {
 		if(JP2_JP2C != box.type) {
+			if (box.length <= 8) return OPJ_FALSE;
 			cio_skip(cio, box.length - 8);
-			jp2_read_boxhdr(cinfo, cio, &box);
+			if( jp2_read_boxhdr(cinfo, cio, &box) == OPJ_FALSE ) return OPJ_FALSE;
 		}
 	} while(JP2_JP2C != box.type);
 
 	*j2k_codestream_offset = cio_tell(cio);
+	if (box.length <= 8) return OPJ_FALSE;
 	*j2k_codestream_length = box.length - 8;
 
 	return OPJ_TRUE;
@@ -930,7 +1041,10 @@ static opj_bool jp2_read_jp(opj_jp2_t *jp2, opj_cio_t *cio) {
 
 	opj_common_ptr cinfo = jp2->cinfo;
 
-	jp2_read_boxhdr(cinfo, cio, &box);
+  if( jp2_read_boxhdr(cinfo, cio, &box) == OPJ_FALSE ) {
+    opj_event_msg(cinfo, EVT_ERROR, "Failed to read boxhdr\n");
+    return OPJ_FALSE;
+  }
 	if (JP2_JP != box.type) {
 		opj_event_msg(cinfo, EVT_ERROR, "Expected JP Marker\n");
 		return OPJ_FALSE;
@@ -1070,6 +1184,7 @@ void jp2_setup_decoder(opj_jp2_t *jp2, opj_dparameters_t *parameters) {
 opj_jp2_t* jp2_create_compress(opj_common_ptr cinfo) {
 	opj_jp2_t *jp2 = (opj_jp2_t*)opj_malloc(sizeof(opj_jp2_t));
 	if(jp2) {
+		memset(jp2, 0, sizeof(opj_jp2_t));
 		jp2->cinfo = cinfo;
 		/* create the J2K codec */
 		jp2->j2k = j2k_create_compress(cinfo);

extern/libopenjpeg/openjpeg.h

@@ -135,6 +135,10 @@ typedef enum COLOR_SPACE {
 	CLRSPC_SYCC = 3			/**< YUV */
 } OPJ_COLOR_SPACE;
 
+#define ENUMCS_SRGB 16
+#define ENUMCS_GRAY 17
+#define ENUMCS_SYCC 18
+
 /**
 Supported codec
 */

extern/libopenjpeg/opj_config.h

@@ -3,7 +3,7 @@
  * the endian check is a blender define */
 
 /* create config.h for CMake */
-#define PACKAGE_VERSION "1.5.0"
+#define PACKAGE_VERSION "1.5.2"
 
 #define HAVE_INTTYPES_H
 #define HAVE_MEMORY_H
@@ -23,6 +23,7 @@
 /* #undef HAVE_LIBLCMS2 */
 /* #undef HAVE_LCMS1_H */
 /* #undef HAVE_LCMS2_H */
+/* #undef USE_SYSTEM_GETOPT */
 
 /* Byte order.  */
 /* All compilers that support Mac OS X define either __BIG_ENDIAN__ or

extern/libopenjpeg/opj_includes.h

@@ -89,18 +89,17 @@ Most compilers implement their own version of this keyword ...
 /* MSVC and Borland C do not have lrintf */
 #if defined(_MSC_VER) || defined(__BORLANDC__)
 static INLINE long lrintf(float f){
-#ifdef _M_X64
-    return (long)((f>0.0f) ? (f + 0.5f):(f -0.5f));
-#else
-    int i;
+#ifdef _M_IX86
+    long int i;
  
     _asm{
         fld f
         fistp i
     };
- 
     return i;
-#endif
+#else
+    return (long)((f>0.0f) ? (f + 0.5f):(f -0.5f));
+#endif /* _M_IX86 */
 }
 #endif
 

extern/libopenjpeg/opj_malloc.h

@@ -83,8 +83,10 @@ Allocate memory aligned to a 16 byte boundry
 #else /* Not _WIN32 */
 	#if defined(__sun)
 		#define HAVE_MEMALIGN
+  #elif defined(__FreeBSD__)
+    #define HAVE_POSIX_MEMALIGN
 	/* Linux x86_64 and OSX always align allocations to 16 bytes */
-	#elif !defined(__amd64__) && !defined(__APPLE__)	
+	#elif !defined(__amd64__) && !defined(__APPLE__) && !defined(_AIX)
 		#define HAVE_MEMALIGN
 		#include <malloc.h>			
 	#endif

extern/libopenjpeg/patches/fbsd.patch

@@ -1,13 +0,0 @@
-Index: extern/libopenjpeg/opj_malloc.h
-===================================================================
---- extern/libopenjpeg/opj_malloc.h	(revision 27736)
-+++ extern/libopenjpeg/opj_malloc.h	(working copy)
-@@ -76,7 +76,7 @@
- 	#if defined(__sun)
- 			#define HAVE_MEMALIGN
- 		#elif defined(__GNUC__)
--                        #ifndef __APPLE__
-+                        #if !defined(__APPLE__) && !defined(__FreeBSD__)
- 			    #define HAVE_MEMALIGN
- 			    #include <malloc.h>
-                         #endif

extern/libopenjpeg/patches/osx.patch

@@ -1,17 +0,0 @@
-Index: opj_malloc.h
-===================================================================
---- opj_malloc.h	(revision 15089)
-+++ opj_malloc.h	(working copy)
-@@ -76,8 +76,10 @@
- 	#if defined(__sun)
- 			#define HAVE_MEMALIGN
- 		#elif defined(__GNUC__)
--			#define HAVE_MEMALIGN
--			#include <malloc.h>		
-+                        #ifndef __APPLE__
-+			    #define HAVE_MEMALIGN
-+			    #include <malloc.h>
-+                        #endif
- 		/* Linux x86_64 and OSX always align allocations to 16 bytes */
- 		#elif !defined(__amd64__) && !defined(__APPLE__)	
- 			/* FIXME: Yes, this is a big assumption */

extern/libopenjpeg/t1.c

@@ -1577,6 +1577,7 @@ void t1_decode_cblks(
 					opj_free(cblk->segs);
 				} /* cblkno */
 				opj_free(precinct->cblks.dec);
+                precinct->cblks.dec = NULL;
 			} /* precno */
 		} /* bandno */
 	} /* resno */

extern/libopenjpeg/t2.c

@@ -30,6 +30,7 @@
  */
 
 #include "opj_includes.h"
+#include <assert.h>
 
 /** @defgroup T2 T2 - Implementation of a tier-2 coding */
 /*@{*/
@@ -64,7 +65,7 @@ static int t2_encode_packet(opj_tcd_tile_t *tile, opj_tcp_t *tcp, opj_pi_iterato
 @param cblksty
 @param first
 */
-static void t2_init_seg(opj_tcd_cblk_dec_t* cblk, int index, int cblksty, int first);
+static opj_bool t2_init_seg(opj_tcd_cblk_dec_t* cblk, int index, int cblksty, int first);
 /**
 Decode a packet of a tile from a source buffer
 @param t2 T2 handle
@@ -296,9 +297,17 @@ static int t2_encode_packet(opj_tcd_tile_t * tile, opj_tcp_t * tcp, opj_pi_itera
 	return (c - dest);
 }
 
-static void t2_init_seg(opj_tcd_cblk_dec_t* cblk, int index, int cblksty, int first) {
+static opj_bool t2_init_seg(opj_tcd_cblk_dec_t* cblk, int index, int cblksty, int first) {
 	opj_tcd_seg_t* seg;
-	cblk->segs = (opj_tcd_seg_t*) opj_realloc(cblk->segs, (index + 1) * sizeof(opj_tcd_seg_t));
+    opj_tcd_seg_t* segs;
+    segs = (opj_tcd_seg_t*) opj_realloc(cblk->segs, (index + 1) * sizeof(opj_tcd_seg_t));
+
+    if (segs == NULL)
+    {
+        return OPJ_FALSE;
+    }
+    cblk->segs = segs;
+
 	seg = &cblk->segs[index];
 	seg->data = NULL;
 	seg->dataindex = 0;
@@ -316,6 +325,8 @@ static void t2_init_seg(opj_tcd_cblk_dec_t* cblk, int index, int cblksty, int fi
 	} else {
 		seg->maxpasses = 109;
 	}
+
+    return OPJ_TRUE;
 }
 
 static int t2_decode_packet(opj_t2_t* t2, unsigned char *src, int len, opj_tcd_tile_t *tile, 
@@ -330,13 +341,15 @@ static int t2_decode_packet(opj_t2_t* t2, unsigned char *src, int len, opj_tcd_t
 	int precno = pi->precno;	/* precinct value */
 	int layno  = pi->layno;		/* quality layer value */
 
-	opj_tcd_resolution_t* res = &tile->comps[compno].resolutions[resno];
-
 	unsigned char *hd = NULL;
 	int present;
 	
 	opj_bio_t *bio = NULL;	/* BIO component */
-	
+
+	opj_tcd_resolution_t* res;
+	assert(&tile->comps[compno] != NULL);
+	res = &tile->comps[compno].resolutions[resno];
+
 	if (layno == 0) {
 		for (bandno = 0; bandno < res->numbands; bandno++) {
 			opj_tcd_band_t *band = &res->bands[bandno];
@@ -462,12 +475,22 @@ static int t2_decode_packet(opj_t2_t* t2, unsigned char *src, int len, opj_tcd_t
 			cblk->numlenbits += increment;
 			segno = 0;
 			if (!cblk->numsegs) {
-				t2_init_seg(cblk, segno, tcp->tccps[compno].cblksty, 1);
+                if (!t2_init_seg(cblk, segno, tcp->tccps[compno].cblksty, 1))
+                {
+                    opj_event_msg(t2->cinfo, EVT_ERROR, "Out of memory\n");
+                    bio_destroy(bio);
+                    return -999;
+                }
 			} else {
 				segno = cblk->numsegs - 1;
 				if (cblk->segs[segno].numpasses == cblk->segs[segno].maxpasses) {
 					++segno;
-					t2_init_seg(cblk, segno, tcp->tccps[compno].cblksty, 0);
+                    if (!t2_init_seg(cblk, segno, tcp->tccps[compno].cblksty, 0))
+                    {
+                        opj_event_msg(t2->cinfo, EVT_ERROR, "Out of memory\n");
+                        bio_destroy(bio);
+                        return -999;
+                    }
 				}
 			}
 			n = cblk->numnewpasses;
@@ -478,7 +501,12 @@ static int t2_decode_packet(opj_t2_t* t2, unsigned char *src, int len, opj_tcd_t
 				n -= cblk->segs[segno].numnewpasses;
 				if (n > 0) {
 					++segno;
-					t2_init_seg(cblk, segno, tcp->tccps[compno].cblksty, 0);
+                    if (!t2_init_seg(cblk, segno, tcp->tccps[compno].cblksty, 0))
+                    {
+                        opj_event_msg(t2->cinfo, EVT_ERROR, "Out of memory\n");
+                        bio_destroy(bio);
+                        return -999;
+                    }
 				}
 			} while (n > 0);
 		}
@@ -714,7 +742,11 @@ int t2_decode_packets(opj_t2_t *t2, unsigned char *src, int len, int tileno, opj
 			} else {
 				e = 0;
 			}
-			if(e == -999) return -999;
+            if(e == -999)
+            {
+                pi_destroy(pi, cp, tileno);
+                return -999;
+            }
 			/* progression in resolution */
 			image->comps[pi[pino].compno].resno_decoded =	
 				(e > 0) ? 

extern/libopenjpeg/tcd.c

@@ -30,7 +30,9 @@
  * POSSIBILITY OF SUCH DAMAGE.
  */
 
+#define _ISOC99_SOURCE /* lrintf is C99 */
 #include "opj_includes.h"
+#include <assert.h>
 
 void tcd_dump(FILE *fd, opj_tcd_t *tcd, opj_tcd_image_t * img) {
 	int tileno, compno, resno, bandno, precno;/*, cblkno;*/
@@ -249,7 +251,9 @@ void tcd_malloc_encode(opj_tcd_t *tcd, opj_image_t * image, opj_cp_t * cp, int c
 					cbgwidthexpn = pdx - 1;
 					cbgheightexpn = pdy - 1;
 				}
-				
+        (void)brcbgyend;
+        (void)brcbgxend;
+
 				cblkwidthexpn = int_min(tccp->cblkw, cbgwidthexpn);
 				cblkheightexpn = int_min(tccp->cblkh, cbgheightexpn);
 				
@@ -333,8 +337,10 @@ void tcd_malloc_encode(opj_tcd_t *tcd, opj_image_t * image, opj_cp_t * cp, int c
 							cblk->y0 = int_max(cblkystart, prc->y0);
 							cblk->x1 = int_min(cblkxend, prc->x1);
 							cblk->y1 = int_min(cblkyend, prc->y1);
-							cblk->data = (unsigned char*) opj_calloc(8192+2, sizeof(unsigned char));
+							cblk->data = (unsigned char*) opj_calloc(9728+2, sizeof(unsigned char));
 							/* FIXME: mqc_init_enc and mqc_byteout underrun the buffer if we don't do this. Why? */
+							cblk->data[0] = 0;
+							cblk->data[1] = 0;
 							cblk->data += 2;
 							cblk->layers = (opj_tcd_layer_t*) opj_calloc(100, sizeof(opj_tcd_layer_t));
 							cblk->passes = (opj_tcd_pass_t*) opj_calloc(100, sizeof(opj_tcd_pass_t));
@@ -509,6 +515,8 @@ void tcd_init_encode(opj_tcd_t *tcd, opj_image_t * image, opj_cp_t * cp, int cur
 					cbgwidthexpn = pdx - 1;
 					cbgheightexpn = pdy - 1;
 				}
+        (void)brcbgyend;
+        (void)brcbgxend;
 				
 				cblkwidthexpn = int_min(tccp->cblkw, cbgwidthexpn);
 				cblkheightexpn = int_min(tccp->cblkh, cbgheightexpn);
@@ -594,6 +602,8 @@ void tcd_init_encode(opj_tcd_t *tcd, opj_image_t * image, opj_cp_t * cp, int cur
 							cblk->y1 = int_min(cblkyend, prc->y1);
 							cblk->data = (unsigned char*) opj_calloc(8192+2, sizeof(unsigned char));
 							/* FIXME: mqc_init_enc and mqc_byteout underrun the buffer if we don't do this. Why? */
+							cblk->data[0] = 0;
+							cblk->data[1] = 0;
 							cblk->data += 2;
 							cblk->layers = (opj_tcd_layer_t*) opj_calloc(100, sizeof(opj_tcd_layer_t));
 							cblk->passes = (opj_tcd_pass_t*) opj_calloc(100, sizeof(opj_tcd_pass_t));
@@ -614,7 +624,7 @@ void tcd_malloc_decode(opj_tcd_t *tcd, opj_image_t * image, opj_cp_t * cp) {
 	tcd->image = image;
 	tcd->tcd_image->tw = cp->tw;
 	tcd->tcd_image->th = cp->th;
-	tcd->tcd_image->tiles = (opj_tcd_tile_t *) opj_malloc(cp->tw * cp->th * sizeof(opj_tcd_tile_t));
+    tcd->tcd_image->tiles = (opj_tcd_tile_t *) opj_calloc(cp->tw * cp->th, sizeof(opj_tcd_tile_t));
 
 	/* 
 	Allocate place to store the decoded data = final image
@@ -690,6 +700,12 @@ void tcd_malloc_decode_tile(opj_tcd_t *tcd, opj_image_t * image, opj_cp_t * cp,
 		opj_tccp_t *tccp = &tcp->tccps[compno];
 		opj_tcd_tilecomp_t *tilec = &tile->comps[compno];
 		
+		if (tccp->numresolutions <= 0)
+		{
+			cp->tileno[tileno] = -1;
+			return;
+		}
+
 		/* border of each tile component (global) */
 		tilec->x0 = int_ceildiv(tile->x0, image->comps[compno].dx);
 		tilec->y0 = int_ceildiv(tile->y0, image->comps[compno].dy);
@@ -749,6 +765,8 @@ void tcd_malloc_decode_tile(opj_tcd_t *tcd, opj_image_t * image, opj_cp_t * cp,
 				cbgwidthexpn = pdx - 1;
 				cbgheightexpn = pdy - 1;
 			}
+      (void)brcbgyend;
+      (void)brcbgxend;
 			
 			cblkwidthexpn = int_min(tccp->cblkw, cbgwidthexpn);
 			cblkheightexpn = int_min(tccp->cblkh, cbgheightexpn);
@@ -1370,16 +1388,31 @@ opj_bool tcd_decode_tile(opj_tcd_t *tcd, unsigned char *src, int len, int tileno
 	if (l == -999) {
 		eof = 1;
 		opj_event_msg(tcd->cinfo, EVT_ERROR, "tcd_decode: incomplete bistream\n");
+    return OPJ_FALSE;
 	}
 	
 	/*------------------TIER1-----------------*/
 	
 	t1_time = opj_clock();	/* time needed to decode a tile */
 	t1 = t1_create(tcd->cinfo);
+    if (t1 == NULL)
+    {
+        opj_event_msg(tcd->cinfo, EVT_ERROR, "Out of memory\n");
+        t1_destroy(t1);
+        return OPJ_FALSE;
+    }
+
 	for (compno = 0; compno < tile->numcomps; ++compno) {
 		opj_tcd_tilecomp_t* tilec = &tile->comps[compno];
 		/* The +3 is headroom required by the vectorized DWT */
 		tilec->data = (int*) opj_aligned_malloc((((tilec->x1 - tilec->x0) * (tilec->y1 - tilec->y0))+3) * sizeof(int));
+        if (tilec->data == NULL)
+        {
+            opj_event_msg(tcd->cinfo, EVT_ERROR, "Out of memory\n");
+            t1_destroy(t1);
+            return OPJ_FALSE;
+        }
+
 		t1_decode_cblks(t1, tilec, &tcd->tcp->tccps[compno]);
 	}
 	t1_destroy(t1);
@@ -1394,13 +1427,15 @@ opj_bool tcd_decode_tile(opj_tcd_t *tcd, unsigned char *src, int len, int tileno
 		int numres2decode;
 
 		if (tcd->cp->reduce != 0) {
-			tcd->image->comps[compno].resno_decoded =
-				tile->comps[compno].numresolutions - tcd->cp->reduce - 1;
-			if (tcd->image->comps[compno].resno_decoded < 0) {				
+			if ( tile->comps[compno].numresolutions < ( tcd->cp->reduce - 1 ) ) {				
 				opj_event_msg(tcd->cinfo, EVT_ERROR, "Error decoding tile. The number of resolutions to remove [%d+1] is higher than the number "
 					" of resolutions in the original codestream [%d]\nModify the cp_reduce parameter.\n", tcd->cp->reduce, tile->comps[compno].numresolutions);
 				return OPJ_FALSE;
 			}
+      else {
+		  	tcd->image->comps[compno].resno_decoded =
+				tile->comps[compno].numresolutions - tcd->cp->reduce - 1;
+      }
 		}
 
 		numres2decode = tcd->image->comps[compno].resno_decoded + 1;
@@ -1421,6 +1456,13 @@ opj_bool tcd_decode_tile(opj_tcd_t *tcd, unsigned char *src, int len, int tileno
 		int n = (tile->comps[0].x1 - tile->comps[0].x0) * (tile->comps[0].y1 - tile->comps[0].y0);
 
 		if (tile->numcomps >= 3 ){
+      /* testcase 1336.pdf.asan.47.376 */
+      if ((tile->comps[0].x1 - tile->comps[0].x0) * (tile->comps[0].y1 - tile->comps[0].y0) < n ||
+        (  tile->comps[1].x1 - tile->comps[1].x0) * (tile->comps[1].y1 - tile->comps[1].y0) < n ||
+        (  tile->comps[2].x1 - tile->comps[2].x0) * (tile->comps[2].y1 - tile->comps[2].y0) < n) {
+        opj_event_msg(tcd->cinfo, EVT_ERROR, "Tiles don't all have the same dimension. Skip the MCT step.\n");
+        return OPJ_FALSE;
+      }
 			if (tcd->tcp->tccps[0].qmfbid == 1) {
 				mct_decode(
 						tile->comps[0].data,
@@ -1452,18 +1494,33 @@ opj_bool tcd_decode_tile(opj_tcd_t *tcd, unsigned char *src, int len, int tileno
 		int tw = tilec->x1 - tilec->x0;
 		int w = imagec->w;
 
+		int i, j;
 		int offset_x = int_ceildivpow2(imagec->x0, imagec->factor);
 		int offset_y = int_ceildivpow2(imagec->y0, imagec->factor);
+    /* NR-DEC-2977.pdf.asan.67.2198.jp2-52-decode */
+    if( res->x0 - offset_x < 0 || res->x1 - offset_x < 0
+     || res->y0 - offset_y < 0 || res->y1 - offset_y < 0 )
+      {
+      opj_event_msg(tcd->cinfo, EVT_ERROR, "Impossible offsets %d / %d\n", offset_x, offset_y);
+      return OPJ_FALSE;
+      }
+    assert( 0 <= res->x0 - offset_x && 0 <= res->x1 - offset_x );
+    assert( 0 <= res->y0 - offset_y && 0 <= res->y1 - offset_y );
 
-		int i, j;
 		if(!imagec->data){
 			imagec->data = (int*) opj_malloc(imagec->w * imagec->h * sizeof(int));
 		}
+        if (!imagec->data)
+        {
+            opj_event_msg(tcd->cinfo, EVT_ERROR, "Out of memory\n");
+            return OPJ_FALSE;
+        }
 		if(tcd->tcp->tccps[compno].qmfbid == 1) {
 			for(j = res->y0; j < res->y1; ++j) {
 				for(i = res->x0; i < res->x1; ++i) {
 					int v = tilec->data[i - res->x0 + (j - res->y0) * tw];
 					v += adjust;
+          /*assert( (i - offset_x) + (j - offset_y) * w >= 0 );*/
 					imagec->data[(i - offset_x) + (j - offset_y) * w] = int_clamp(v, min, max);
 				}
 			}
@@ -1473,6 +1530,7 @@ opj_bool tcd_decode_tile(opj_tcd_t *tcd, unsigned char *src, int len, int tileno
 					float tmp = ((float*)tilec->data)[i - res->x0 + (j - res->y0) * tw];
 					int v = lrintf(tmp);
 					v += adjust;
+          /*assert( (i - offset_x) + (j - offset_y) * w >= 0 );*/
 					imagec->data[(i - offset_x) + (j - offset_y) * w] = int_clamp(v, min, max);
 				}
 			}
@@ -1492,32 +1550,51 @@ opj_bool tcd_decode_tile(opj_tcd_t *tcd, unsigned char *src, int len, int tileno
 
 void tcd_free_decode(opj_tcd_t *tcd) {
 	opj_tcd_image_t *tcd_image = tcd->tcd_image;	
+    int i = 0;
+    for (i = 0; i < tcd_image->tw * tcd_image->th; i++)
+    {
+        tcd_free_decode_tile(tcd, i);
+    }
+
 	opj_free(tcd_image->tiles);
 }
 
 void tcd_free_decode_tile(opj_tcd_t *tcd, int tileno) {
-	int compno,resno,bandno,precno;
+    int compno,resno,bandno,precno,cblkno;
 
 	opj_tcd_image_t *tcd_image = tcd->tcd_image;
 
 	opj_tcd_tile_t *tile = &tcd_image->tiles[tileno];
-	for (compno = 0; compno < tile->numcomps; compno++) {
-		opj_tcd_tilecomp_t *tilec = &tile->comps[compno];
-		for (resno = 0; resno < tilec->numresolutions; resno++) {
-			opj_tcd_resolution_t *res = &tilec->resolutions[resno];
-			for (bandno = 0; bandno < res->numbands; bandno++) {
-				opj_tcd_band_t *band = &res->bands[bandno];
-				for (precno = 0; precno < res->ph * res->pw; precno++) {
-					opj_tcd_precinct_t *prec = &band->precincts[precno];
-					if (prec->imsbtree != NULL) tgt_destroy(prec->imsbtree);
-					if (prec->incltree != NULL) tgt_destroy(prec->incltree);
+    if (tile->comps != NULL) {
+        for (compno = 0; compno < tile->numcomps; compno++) {
+            opj_tcd_tilecomp_t *tilec = &tile->comps[compno];
+            for (resno = 0; resno < tilec->numresolutions; resno++) {
+                opj_tcd_resolution_t *res = &tilec->resolutions[resno];
+                for (bandno = 0; bandno < res->numbands; bandno++) {
+                    opj_tcd_band_t *band = &res->bands[bandno];
+                    for (precno = 0; precno < res->ph * res->pw; precno++) {
+                        opj_tcd_precinct_t *prec = &band->precincts[precno];
+                        if (prec->cblks.dec != NULL) {
+                            for (cblkno = 0; cblkno < prec->cw * prec->ch; ++cblkno) {
+                                opj_tcd_cblk_dec_t* cblk = &prec->cblks.dec[cblkno];
+                                opj_free(cblk->data);
+                                opj_free(cblk->segs);
+                            }
+                            opj_free(prec->cblks.dec);
+                        }
+                        if (prec->imsbtree != NULL) tgt_destroy(prec->imsbtree);
+                        if (prec->incltree != NULL) tgt_destroy(prec->incltree);
+
+
+                    }
+                    opj_free(band->precincts);
 				}
-				opj_free(band->precincts);
 			}
+            opj_free(tilec->resolutions);
 		}
-		opj_free(tilec->resolutions);
+        opj_free(tile->comps);
+        tile->comps = NULL;
 	}
-	opj_free(tile->comps);
 }
 
 

extern/lzma/LzmaEnc.c

@@ -1919,11 +1919,10 @@ static SRes LzmaEnc_CodeOneBlock(CLzmaEnc *p, Bool useLimits, UInt32 maxPackSize
 static SRes LzmaEnc_Alloc(CLzmaEnc *p, UInt32 keepWindowSize, ISzAlloc *alloc, ISzAlloc *allocBig)
 {
   UInt32 beforeSize = kNumOpts;
-  Bool btMode;
   if (!RangeEnc_Alloc(&p->rc, alloc))
     return SZ_ERROR_MEM;
-  btMode = (p->matchFinderBase.btMode != 0);
   #ifdef COMPRESS_MF_MT
+  Bool btMode = (p->matchFinderBase.btMode != 0);;
   p->mtMode = (p->multiThread && !p->fastMode && btMode);
   #endif
 

extern/recastnavigation/CMakeLists.txt

@@ -23,9 +23,11 @@
 #
 # ***** END GPL LICENSE BLOCK *****
 
+remove_cc_flag_unsigned_char()
+
 set(INC 
-		Recast/Include
-		Detour/Include
+	Recast/Include
+	Detour/Include
 )
 
 set(INC_SYS
@@ -33,38 +35,38 @@ set(INC_SYS
 )
 
 set(SRC 
-		recast-capi.cpp
-		recast-capi.h
+	recast-capi.cpp
+	recast-capi.h
 
 
-		Detour/Source/DetourCommon.cpp
-		Detour/Source/DetourNode.cpp
-		Detour/Source/DetourStatNavMesh.cpp
-		Detour/Source/DetourStatNavMeshBuilder.cpp
-		Detour/Source/DetourTileNavMesh.cpp
-		Detour/Source/DetourTileNavMeshBuilder.cpp
+	Detour/Source/DetourCommon.cpp
+	Detour/Source/DetourNode.cpp
+	Detour/Source/DetourStatNavMesh.cpp
+	Detour/Source/DetourStatNavMeshBuilder.cpp
+	Detour/Source/DetourTileNavMesh.cpp
+	Detour/Source/DetourTileNavMeshBuilder.cpp
 
-		Detour/Include/DetourCommon.h
-		Detour/Include/DetourNode.h
-		Detour/Include/DetourStatNavMesh.h
-		Detour/Include/DetourStatNavMeshBuilder.h
-		Detour/Include/DetourTileNavMesh.h
-		Detour/Include/DetourTileNavMeshBuilder.h
+	Detour/Include/DetourCommon.h
+	Detour/Include/DetourNode.h
+	Detour/Include/DetourStatNavMesh.h
+	Detour/Include/DetourStatNavMeshBuilder.h
+	Detour/Include/DetourTileNavMesh.h
+	Detour/Include/DetourTileNavMeshBuilder.h
 
-		Recast/Source/Recast.cpp
-		Recast/Source/RecastAlloc.cpp
-		Recast/Source/RecastArea.cpp
-		Recast/Source/RecastContour.cpp
-		Recast/Source/RecastFilter.cpp
-		Recast/Source/RecastLayers.cpp
-		Recast/Source/RecastMesh.cpp
-		Recast/Source/RecastMeshDetail.cpp
-		Recast/Source/RecastRasterization.cpp
-		Recast/Source/RecastRegion.cpp
+	Recast/Source/Recast.cpp
+	Recast/Source/RecastAlloc.cpp
+	Recast/Source/RecastArea.cpp
+	Recast/Source/RecastContour.cpp
+	Recast/Source/RecastFilter.cpp
+	Recast/Source/RecastLayers.cpp
+	Recast/Source/RecastMesh.cpp
+	Recast/Source/RecastMeshDetail.cpp
+	Recast/Source/RecastRasterization.cpp
+	Recast/Source/RecastRegion.cpp
 
-		Recast/Include/Recast.h
-		Recast/Include/RecastAlloc.h
-		Recast/Include/RecastAssert.h
+	Recast/Include/Recast.h
+	Recast/Include/RecastAlloc.h
+	Recast/Include/RecastAssert.h
 )
 
 blender_add_lib(extern_recastnavigation "${SRC}" "${INC}" "${INC_SYS}")

extern/recastnavigation/Recast/Include/RecastAlloc.h

@@ -19,6 +19,8 @@
 #ifndef RECASTALLOC_H
 #define RECASTALLOC_H
 
+#include <stddef.h>
+
 /// Provides hint values to the memory allocator on how long the
 /// memory is expected to be used.
 enum rcAllocHint
@@ -28,30 +30,32 @@ enum rcAllocHint
 };
 
 /// A memory allocation function.
-//  @param[in] size         The size, in bytes of memory, to allocate.
-//  @param[in] rcAllocHint  A hint to the allocator on how long the memory is expected to be in use.
+//  @param[in]		size			The size, in bytes of memory, to allocate.
+//  @param[in]		rcAllocHint	A hint to the allocator on how long the memory is expected to be in use.
 //  @return A pointer to the beginning of the allocated memory block, or null if the allocation failed.
 ///  @see rcAllocSetCustom
-typedef void* (rcAllocFunc)(int size, rcAllocHint hint);
+typedef void* (rcAllocFunc)(size_t size, rcAllocHint hint);
 
 /// A memory deallocation function.
+///  @param[in]		ptr		A pointer to a memory block previously allocated using #rcAllocFunc.
 /// @see rcAllocSetCustom
-//  @param[in] ptr 
 typedef void (rcFreeFunc)(void* ptr);
 
 /// Sets the base custom allocation functions to be used by Recast.
-///  @param[in] allocFunc  The memory allocation function to be used by #rcAlloc
-///  @param[in] freeFunc   The memory de-allocation function to be used by #rcFree
+///  @param[in]		allocFunc	The memory allocation function to be used by #rcAlloc
+///  @param[in]		freeFunc	The memory de-allocation function to be used by #rcFree
 void rcAllocSetCustom(rcAllocFunc *allocFunc, rcFreeFunc *freeFunc);
 
 /// Allocates a memory block.
-///  @param[in] size  The size, in bytes of memory, to allocate.
-///  @param[in] hint  A hint to the allocator on how long the memory is expected to be in use.
+///  @param[in]		size	The size, in bytes of memory, to allocate.
+///  @param[in]		hint	A hint to the allocator on how long the memory is expected to be in use.
 ///  @return A pointer to the beginning of the allocated memory block, or null if the allocation failed.
-void* rcAlloc(int size, rcAllocHint hint);
+/// @see rcFree
+void* rcAlloc(size_t size, rcAllocHint hint);
 
 /// Deallocates a memory block.
-///  @param[in] ptr A pointer to a memory block previously allocated using #rcAlloc.
+///  @param[in]		ptr		A pointer to a memory block previously allocated using #rcAlloc.
+/// @see rcAlloc
 void rcFree(void* ptr);
 
 
@@ -60,42 +64,58 @@ class rcIntArray
 {
 	int* m_data;
 	int m_size, m_cap;
-	inline rcIntArray(const rcIntArray&);
-	inline rcIntArray& operator=(const rcIntArray&);
-public:
 
+	void doResize(int n);
+	
+	// Explicitly disabled copy constructor and copy assignment operator.
+	rcIntArray(const rcIntArray&);
+	rcIntArray& operator=(const rcIntArray&);
+
+public:
 	/// Constructs an instance with an initial array size of zero.
-	inline rcIntArray() : m_data(0), m_size(0), m_cap(0) {}
+	rcIntArray() : m_data(0), m_size(0), m_cap(0) {}
 
 	/// Constructs an instance initialized to the specified size.
-	///  @param[in] n The initial size of the integer array.
-	inline rcIntArray(int n) : m_data(0), m_size(0), m_cap(0) { resize(n); }
-	inline ~rcIntArray() { rcFree(m_data); }
+	///  @param[in]		n	The initial size of the integer array.
+	rcIntArray(int n) : m_data(0), m_size(0), m_cap(0) { resize(n); }
+	~rcIntArray() { rcFree(m_data); }
 
 	/// Specifies the new size of the integer array.
-	///  @param[in] n  The new size of the integer array.
-	void resize(int n);
+	///  @param[in]		n	The new size of the integer array.
+	void resize(int n)
+	{
+		if (n > m_cap)
+			doResize(n);
+		
+		m_size = n;
+	}
 
 	/// Push the specified integer onto the end of the array and increases the size by one.
-	///  @param[in] item  The new value.
-	inline void push(int item) { resize(m_size+1); m_data[m_size-1] = item; }
+	///  @param[in]		item	The new value.
+	void push(int item) { resize(m_size+1); m_data[m_size-1] = item; }
 
 	/// Returns the value at the end of the array and reduces the size by one.
 	///  @return The value at the end of the array.
-	inline int pop() { if (m_size > 0) m_size--; return m_data[m_size]; }
+	int pop()
+	{
+		if (m_size > 0)
+			m_size--;
+		
+		return m_data[m_size];
+	}
 
 	/// The value at the specified array index.
 	/// @warning Does not provide overflow protection.
-	///  @param[in] i  The index of the value.
-	inline const int& operator[](int i) const { return m_data[i]; }
+	///  @param[in]		i	The index of the value.
+	const int& operator[](int i) const { return m_data[i]; }
 
 	/// The value at the specified array index.
 	/// @warning Does not provide overflow protection.
-	///  @param[in] i  The index of the value.
-	inline int& operator[](int i) { return m_data[i]; }
+	///  @param[in]		i	The index of the value.
+	int& operator[](int i) { return m_data[i]; }
 
 	/// The current size of the integer array.
-	inline int size() const { return m_size; }
+	int size() const { return m_size; }
 };
 
 /// A simple helper class used to delete an array when it goes out of scope.
@@ -110,13 +130,18 @@ public:
 	inline rcScopedDelete() : ptr(0) {}
 
 	/// Constructs an instance with the specified pointer.
-	///  @param[in] p  An pointer to an allocated array.
+	///  @param[in]		p	An pointer to an allocated array.
 	inline rcScopedDelete(T* p) : ptr(p) {}
 	inline ~rcScopedDelete() { rcFree(ptr); }
 
 	/// The root array pointer.
 	///  @return The root array pointer.
 	inline operator T*() { return ptr; }
+	
+private:
+	// Explicitly disabled copy constructor and copy assignment operator.
+	rcScopedDelete(const rcScopedDelete&);
+	rcScopedDelete& operator=(const rcScopedDelete&);
 };
 
 #endif

extern/recastnavigation/Recast/Include/RecastAssert.h

@@ -24,7 +24,7 @@
 
 #ifdef NDEBUG
 // From http://cnicholson.net/2009/02/stupid-c-tricks-adventures-in-assert/
-#	define rcAssert(x) do { (void)sizeof(x); } while((void)(__LINE__ == -1), false)
+#	define rcAssert(x) do { (void)sizeof(x); } while((void)(__LINE__==-1),false)  
 #else
 #	include <assert.h> 
 #	define rcAssert assert

extern/recastnavigation/Recast/Include/RecastLog.h

@@ -1,80 +0,0 @@
-//
-// Copyright (c) 2009 Mikko Mononen memon@inside.org
-//
-// This software is provided 'as-is', without any express or implied
-// warranty.  In no event will the authors be held liable for any damages
-// arising from the use of this software.
-// Permission is granted to anyone to use this software for any purpose,
-// including commercial applications, and to alter it and redistribute it
-// freely, subject to the following restrictions:
-// 1. The origin of this software must not be misrepresented; you must not
-//    claim that you wrote the original software. If you use this software
-//    in a product, an acknowledgment in the product documentation would be
-//    appreciated but is not required.
-// 2. Altered source versions must be plainly marked as such, and must not be
-//    misrepresented as being the original software.
-// 3. This notice may not be removed or altered from any source distribution.
-//
-
-#ifndef RECAST_LOG_H
-#define RECAST_LOG_H
-
-enum rcLogCategory
-{
-	RC_LOG_PROGRESS = 1,
-	RC_LOG_WARNING,
-	RC_LOG_ERROR,
-};
-
-class rcLog
-{
-public:
-	rcLog();
-	~rcLog();
-	
-	void log(rcLogCategory category, const char* format, ...);
-	inline void clear() { m_messageCount = 0; m_textPoolSize = 0; }
-	inline int getMessageCount() const { return m_messageCount; }
-	inline char getMessageType(int i) const { return *m_messages[i]; }
-	inline const char* getMessageText(int i) const { return m_messages[i]+1; }
-
-private:
-	static const int MAX_MESSAGES = 1000;
-	const char* m_messages[MAX_MESSAGES];
-	int m_messageCount;
-	static const int TEXT_POOL_SIZE = 8000;
-	char m_textPool[TEXT_POOL_SIZE];
-	int m_textPoolSize;
-};
-
-struct rcBuildTimes
-{
-	int rasterizeTriangles;
-	int buildCompact;
-	int buildContours;
-	int buildContoursTrace;
-	int buildContoursSimplify;
-	int filterBorder;
-	int filterWalkable;
-	int filterMarkReachable;
-	int buildPolymesh;
-	int buildDistanceField;
-	int buildDistanceFieldDist;
-	int buildDistanceFieldBlur;
-	int buildRegions;
-	int buildRegionsReg;
-	int buildRegionsExp;
-	int buildRegionsFlood;
-	int buildRegionsFilter;
-	int buildDetailMesh;
-	int mergePolyMesh;
-	int mergePolyMeshDetail;
-};
-
-void rcSetLog(rcLog* log);
-rcLog* rcGetLog();
-
-void rcSetBuildTimes(rcBuildTimes* btimes);
-rcBuildTimes* rcGetBuildTimes();
-
-#endif // RECAST_LOG_H

extern/recastnavigation/Recast/Include/RecastTimer.h

@@ -1,31 +0,0 @@
-//
-// Copyright (c) 2009 Mikko Mononen memon@inside.org
-//
-// This software is provided 'as-is', without any express or implied
-// warranty.  In no event will the authors be held liable for any damages
-// arising from the use of this software.
-// Permission is granted to anyone to use this software for any purpose,
-// including commercial applications, and to alter it and redistribute it
-// freely, subject to the following restrictions:
-// 1. The origin of this software must not be misrepresented; you must not
-//    claim that you wrote the original software. If you use this software
-//    in a product, an acknowledgment in the product documentation would be
-//    appreciated but is not required.
-// 2. Altered source versions must be plainly marked as such, and must not be
-//    misrepresented as being the original software.
-// 3. This notice may not be removed or altered from any source distribution.
-//
-#ifndef RECAST_TIMER_H
-#define RECAST_TIMER_H
-
-#ifdef __GNUC__
-#include <stdint.h>
-typedef int64_t rcTimeVal;
-#else
-typedef __int64 rcTimeVal;
-#endif
-
-rcTimeVal rcGetPerformanceTimer();
-int rcGetDeltaTimeUsec(rcTimeVal start, rcTimeVal end);
-
-#endif // RECAST_TIMER_H

extern/recastnavigation/Recast/Source/Recast.cpp

@@ -208,12 +208,11 @@ void rcCalcGridSize(const float* bmin, const float* bmax, float cs, int* w, int*
 /// See the #rcConfig documentation for more information on the configuration parameters.
 /// 
 /// @see rcAllocHeightfield, rcHeightfield 
-bool rcCreateHeightfield(rcContext* /*ctx*/, rcHeightfield& hf, int width, int height,
+bool rcCreateHeightfield(rcContext* ctx, rcHeightfield& hf, int width, int height,
 						 const float* bmin, const float* bmax,
 						 float cs, float ch)
 {
-	// TODO: VC complains about unref formal variable, figure out a way to handle this better.
-//	rcAssert(ctx);
+	rcIgnoreUnused(ctx);
 	
 	hf.width = width;
 	hf.height = height;
@@ -239,19 +238,18 @@ static void calcTriNormal(const float* v0, const float* v1, const float* v2, flo
 
 /// @par
 ///
-/// Only sets the aread id's for the walkable triangles.  Does not alter the
+/// Only sets the area id's for the walkable triangles.  Does not alter the
 /// area id's for unwalkable triangles.
 /// 
 /// See the #rcConfig documentation for more information on the configuration parameters.
 /// 
 /// @see rcHeightfield, rcClearUnwalkableTriangles, rcRasterizeTriangles
-void rcMarkWalkableTriangles(rcContext* /*ctx*/, const float walkableSlopeAngle,
+void rcMarkWalkableTriangles(rcContext* ctx, const float walkableSlopeAngle,
 							 const float* verts, int /*nv*/,
 							 const int* tris, int nt,
 							 unsigned char* areas)
 {
-	// TODO: VC complains about unref formal variable, figure out a way to handle this better.
-//	rcAssert(ctx);
+	rcIgnoreUnused(ctx);
 	
 	const float walkableThr = cosf(walkableSlopeAngle/180.0f*RC_PI);
 
@@ -269,19 +267,18 @@ void rcMarkWalkableTriangles(rcContext* /*ctx*/, const float walkableSlopeAngle,
 
 /// @par
 ///
-/// Only sets the aread id's for the unwalkable triangles.  Does not alter the
+/// Only sets the area id's for the unwalkable triangles.  Does not alter the
 /// area id's for walkable triangles.
 /// 
 /// See the #rcConfig documentation for more information on the configuration parameters.
 /// 
 /// @see rcHeightfield, rcClearUnwalkableTriangles, rcRasterizeTriangles
-void rcClearUnwalkableTriangles(rcContext* /*ctx*/, const float walkableSlopeAngle,
+void rcClearUnwalkableTriangles(rcContext* ctx, const float walkableSlopeAngle,
 								const float* verts, int /*nv*/,
 								const int* tris, int nt,
 								unsigned char* areas)
 {
-	// TODO: VC complains about unref formal variable, figure out a way to handle this better.
-//	rcAssert(ctx);
+	rcIgnoreUnused(ctx);
 	
 	const float walkableThr = cosf(walkableSlopeAngle/180.0f*RC_PI);
 	
@@ -297,10 +294,9 @@ void rcClearUnwalkableTriangles(rcContext* /*ctx*/, const float walkableSlopeAng
 	}
 }
 
-int rcGetHeightFieldSpanCount(rcContext* /*ctx*/, rcHeightfield& hf)
+int rcGetHeightFieldSpanCount(rcContext* ctx, rcHeightfield& hf)
 {
-	// TODO: VC complains about unref formal variable, figure out a way to handle this better.
-//	rcAssert(ctx);
+	rcIgnoreUnused(ctx);
 	
 	const int w = hf.width;
 	const int h = hf.height;
@@ -322,7 +318,7 @@ int rcGetHeightFieldSpanCount(rcContext* /*ctx*/, rcHeightfield& hf)
 /// @par
 ///
 /// This is just the beginning of the process of fully building a compact heightfield.
-/// Various filters may be applied applied, then the distance field and regions built.
+/// Various filters may be applied, then the distance field and regions built.
 /// E.g: #rcBuildDistanceField and #rcBuildRegions
 ///
 /// See the #rcConfig documentation for more information on the configuration parameters.
@@ -333,7 +329,7 @@ bool rcBuildCompactHeightfield(rcContext* ctx, const int walkableHeight, const i
 {
 	rcAssert(ctx);
 	
-	ctx->startTimer(RC_TIMER_BUILD_COMPACTHEIGHTFIELD);
+	rcScopedTimer timer(ctx, RC_TIMER_BUILD_COMPACTHEIGHTFIELD);
 	
 	const int w = hf.width;
 	const int h = hf.height;
@@ -439,13 +435,13 @@ bool rcBuildCompactHeightfield(rcContext* ctx, const int walkableHeight, const i
 						if ((top - bot) >= walkableHeight && rcAbs((int)ns.y - (int)s.y) <= walkableClimb)
 						{
 							// Mark direction as walkable.
-							const int idx = k - (int)nc.index;
-							if (idx < 0 || idx > MAX_LAYERS)
+							const int lidx = k - (int)nc.index;
+							if (lidx < 0 || lidx > MAX_LAYERS)
 							{
-								tooHighNeighbour = rcMax(tooHighNeighbour, idx);
+								tooHighNeighbour = rcMax(tooHighNeighbour, lidx);
 								continue;
 							}
-							rcSetCon(s, dir, idx);
+							rcSetCon(s, dir, lidx);
 							break;
 						}
 					}
@@ -460,8 +456,6 @@ bool rcBuildCompactHeightfield(rcContext* ctx, const int walkableHeight, const i
 		ctx->log(RC_LOG_ERROR, "rcBuildCompactHeightfield: Heightfield has too many layers %d (max: %d)",
 				 tooHighNeighbour, MAX_LAYERS);
 	}
-		
-	ctx->stopTimer(RC_TIMER_BUILD_COMPACTHEIGHTFIELD);
 	
 	return true;
 }
@@ -490,4 +484,4 @@ static int getCompactHeightFieldMemoryusage(const rcCompactHeightfield& chf)
 	size += sizeof(rcCompactCell) * chf.width * chf.height;
 	return size;
 }
-*/
+*/

extern/recastnavigation/Recast/Source/RecastAlloc.cpp

@@ -20,7 +20,7 @@
 #include <string.h>
 #include "RecastAlloc.h"
 
-static void *rcAllocDefault(int size, rcAllocHint)
+static void *rcAllocDefault(size_t size, rcAllocHint)
 {
 	return malloc(size);
 }
@@ -41,7 +41,7 @@ void rcAllocSetCustom(rcAllocFunc *allocFunc, rcFreeFunc *freeFunc)
 }
 
 /// @see rcAllocSetCustom
-void* rcAlloc(int size, rcAllocHint hint)
+void* rcAlloc(size_t size, rcAllocHint hint)
 {
 	return sRecastAllocFunc(size, hint);
 }
@@ -72,17 +72,13 @@ void rcFree(void* ptr)
 /// Using this method ensures the array is at least large enough to hold
 /// the specified number of elements.  This can improve performance by
 /// avoiding auto-resizing during use.
-void rcIntArray::resize(int n)
+void rcIntArray::doResize(int n)
 {
-	if (n > m_cap)
-	{
-		if (!m_cap) m_cap = n;
-		while (m_cap < n) m_cap *= 2;
-		int* newData = (int*)rcAlloc(m_cap*sizeof(int), RC_ALLOC_TEMP);
-		if (m_size && newData) memcpy(newData, m_data, m_size*sizeof(int));
-		rcFree(m_data);
-		m_data = newData;
-	}
-	m_size = n;
+	if (!m_cap) m_cap = n;
+	while (m_cap < n) m_cap *= 2;
+	int* newData = (int*)rcAlloc(m_cap*sizeof(int), RC_ALLOC_TEMP);
+	if (m_size && newData) memcpy(newData, m_data, m_size*sizeof(int));
+	rcFree(m_data);
+	m_data = newData;
 }
 

extern/recastnavigation/Recast/Source/RecastArea.cpp

@@ -41,7 +41,7 @@ bool rcErodeWalkableArea(rcContext* ctx, int radius, rcCompactHeightfield& chf)
 	const int w = chf.width;
 	const int h = chf.height;
 	
-	ctx->startTimer(RC_TIMER_ERODE_AREA);
+	rcScopedTimer timer(ctx, RC_TIMER_ERODE_AREA);
 	
 	unsigned char* dist = (unsigned char*)rcAlloc(sizeof(unsigned char)*chf.spanCount, RC_ALLOC_TEMP);
 	if (!dist)
@@ -75,8 +75,8 @@ bool rcErodeWalkableArea(rcContext* ctx, int radius, rcCompactHeightfield& chf)
 						{
 							const int nx = x + rcGetDirOffsetX(dir);
 							const int ny = y + rcGetDirOffsetY(dir);
-							const int ni = (int)chf.cells[nx+ny*w].index + rcGetCon(s, dir);
-							if (chf.areas[ni] != RC_NULL_AREA)
+							const int nidx = (int)chf.cells[nx+ny*w].index + rcGetCon(s, dir);
+							if (chf.areas[nidx] != RC_NULL_AREA)
 							{
 								nc++;
 							}
@@ -215,8 +215,6 @@ bool rcErodeWalkableArea(rcContext* ctx, int radius, rcCompactHeightfield& chf)
 	
 	rcFree(dist);
 	
-	ctx->stopTimer(RC_TIMER_ERODE_AREA);
-	
 	return true;
 }
 
@@ -245,7 +243,7 @@ bool rcMedianFilterWalkableArea(rcContext* ctx, rcCompactHeightfield& chf)
 	const int w = chf.width;
 	const int h = chf.height;
 	
-	ctx->startTimer(RC_TIMER_MEDIAN_AREA);
+	rcScopedTimer timer(ctx, RC_TIMER_MEDIAN_AREA);
 	
 	unsigned char* areas = (unsigned char*)rcAlloc(sizeof(unsigned char)*chf.spanCount, RC_ALLOC_TEMP);
 	if (!areas)
@@ -306,8 +304,6 @@ bool rcMedianFilterWalkableArea(rcContext* ctx, rcCompactHeightfield& chf)
 	memcpy(chf.areas, areas, sizeof(unsigned char)*chf.spanCount);
 	
 	rcFree(areas);
-
-	ctx->stopTimer(RC_TIMER_MEDIAN_AREA);
 	
 	return true;
 }
@@ -322,7 +318,7 @@ void rcMarkBoxArea(rcContext* ctx, const float* bmin, const float* bmax, unsigne
 {
 	rcAssert(ctx);
 	
-	ctx->startTimer(RC_TIMER_MARK_BOX_AREA);
+	rcScopedTimer timer(ctx, RC_TIMER_MARK_BOX_AREA);
 
 	int minx = (int)((bmin[0]-chf.bmin[0])/chf.cs);
 	int miny = (int)((bmin[1]-chf.bmin[1])/chf.ch);
@@ -357,9 +353,6 @@ void rcMarkBoxArea(rcContext* ctx, const float* bmin, const float* bmax, unsigne
 			}
 		}
 	}
-
-	ctx->stopTimer(RC_TIMER_MARK_BOX_AREA);
-
 }
 
 
@@ -391,7 +384,7 @@ void rcMarkConvexPolyArea(rcContext* ctx, const float* verts, const int nverts,
 {
 	rcAssert(ctx);
 	
-	ctx->startTimer(RC_TIMER_MARK_CONVEXPOLY_AREA);
+	rcScopedTimer timer(ctx, RC_TIMER_MARK_CONVEXPOLY_AREA);
 
 	float bmin[3], bmax[3];
 	rcVcopy(bmin, verts);
@@ -448,10 +441,86 @@ void rcMarkConvexPolyArea(rcContext* ctx, const float* verts, const int nverts,
 			}
 		}
 	}
-
-	ctx->stopTimer(RC_TIMER_MARK_CONVEXPOLY_AREA);
 }
 
+int rcOffsetPoly(const float* verts, const int nverts, const float offset,
+				 float* outVerts, const int maxOutVerts)
+{
+	const float	MITER_LIMIT = 1.20f;
+
+	int n = 0;
+
+	for (int i = 0; i < nverts; i++)
+	{
+		const int a = (i+nverts-1) % nverts;
+		const int b = i;
+		const int c = (i+1) % nverts;
+		const float* va = &verts[a*3];
+		const float* vb = &verts[b*3];
+		const float* vc = &verts[c*3];
+		float dx0 = vb[0] - va[0];
+		float dy0 = vb[2] - va[2];
+		float d0 = dx0*dx0 + dy0*dy0;
+		if (d0 > 1e-6f)
+		{
+			d0 = 1.0f/rcSqrt(d0);
+			dx0 *= d0;
+			dy0 *= d0;
+		}
+		float dx1 = vc[0] - vb[0];
+		float dy1 = vc[2] - vb[2];
+		float d1 = dx1*dx1 + dy1*dy1;
+		if (d1 > 1e-6f)
+		{
+			d1 = 1.0f/rcSqrt(d1);
+			dx1 *= d1;
+			dy1 *= d1;
+		}
+		const float dlx0 = -dy0;
+		const float dly0 = dx0;
+		const float dlx1 = -dy1;
+		const float dly1 = dx1;
+		float cross = dx1*dy0 - dx0*dy1;
+		float dmx = (dlx0 + dlx1) * 0.5f;
+		float dmy = (dly0 + dly1) * 0.5f;
+		float dmr2 = dmx*dmx + dmy*dmy;
+		bool bevel = dmr2 * MITER_LIMIT*MITER_LIMIT < 1.0f;
+		if (dmr2 > 1e-6f)
+		{
+			const float scale = 1.0f / dmr2;
+			dmx *= scale;
+			dmy *= scale;
+		}
+
+		if (bevel && cross < 0.0f)
+		{
+			if (n+2 >= maxOutVerts)
+				return 0;
+			float d = (1.0f - (dx0*dx1 + dy0*dy1))*0.5f;
+			outVerts[n*3+0] = vb[0] + (-dlx0+dx0*d)*offset;
+			outVerts[n*3+1] = vb[1];
+			outVerts[n*3+2] = vb[2] + (-dly0+dy0*d)*offset;
+			n++;
+			outVerts[n*3+0] = vb[0] + (-dlx1-dx1*d)*offset;
+			outVerts[n*3+1] = vb[1];
+			outVerts[n*3+2] = vb[2] + (-dly1-dy1*d)*offset;
+			n++;
+		}
+		else
+		{
+			if (n+1 >= maxOutVerts)
+				return 0;
+			outVerts[n*3+0] = vb[0] - dmx*offset;
+			outVerts[n*3+1] = vb[1];
+			outVerts[n*3+2] = vb[2] - dmy*offset;
+			n++;
+		}
+	}
+	
+	return n;
+}
+
+
 /// @par
 ///
 /// The value of spacial parameters are in world units.
@@ -463,7 +532,7 @@ void rcMarkCylinderArea(rcContext* ctx, const float* pos,
 {
 	rcAssert(ctx);
 	
-	ctx->startTimer(RC_TIMER_MARK_CYLINDER_AREA);
+	rcScopedTimer timer(ctx, RC_TIMER_MARK_CYLINDER_AREA);
 	
 	float bmin[3], bmax[3];
 	bmin[0] = pos[0] - r;
@@ -519,6 +588,4 @@ void rcMarkCylinderArea(rcContext* ctx, const float* pos,
 			}
 		}
 	}
-	
-	ctx->stopTimer(RC_TIMER_MARK_CYLINDER_AREA);
 }

extern/recastnavigation/Recast/Source/RecastContour.cpp

@@ -20,6 +20,7 @@
 #include <math.h>
 #include <string.h>
 #include <stdio.h>
+#include <stdlib.h>
 #include "Recast.h"
 #include "RecastAlloc.h"
 #include "RecastAssert.h"
@@ -36,7 +37,7 @@ static int getCornerHeight(int x, int y, int i, int dir,
 	unsigned int regs[4] = {0,0,0,0};
 	
 	// Combine region and area codes in order to prevent
-	// border vertices which are in between two areas to be removed. 
+	// border vertices which are in between two areas to be removed.
 	regs[0] = chf.spans[i].reg | (chf.areas[i] << 16);
 	
 	if (rcGetCon(s, dir) != RC_NOT_CONNECTED)
@@ -187,27 +188,6 @@ static float distancePtSeg(const int x, const int z,
 						   const int px, const int pz,
 						   const int qx, const int qz)
 {
-/*	float pqx = (float)(qx - px);
-	float pqy = (float)(qy - py);
-	float pqz = (float)(qz - pz);
-	float dx = (float)(x - px);
-	float dy = (float)(y - py);
-	float dz = (float)(z - pz);
-	float d = pqx*pqx + pqy*pqy + pqz*pqz;
-	float t = pqx*dx + pqy*dy + pqz*dz;
-	if (d > 0)
-		t /= d;
-	if (t < 0)
-		t = 0;
-	else if (t > 1)
-		t = 1;
-	
-	dx = px + t*pqx - x;
-	dy = py + t*pqy - y;
-	dz = pz + t*pqz - z;
-	
-	return dx*dx + dy*dy + dz*dz;*/
-
 	float pqx = (float)(qx - px);
 	float pqz = (float)(qz - pz);
 	float dx = (float)(x - px);
@@ -257,13 +237,13 @@ static void simplifyContour(rcIntArray& points, rcIntArray& simplified,
 				simplified.push(points[i*4+2]);
 				simplified.push(i);
 			}
-		}       
+		}
 	}
 	
 	if (simplified.size() == 0)
 	{
 		// If there is no connections at all,
-		// create some initial points for the simplification process. 
+		// create some initial points for the simplification process.
 		// Find lower-left and upper-right vertices of the contour.
 		int llx = points[0];
 		int lly = points[1];
@@ -311,19 +291,19 @@ static void simplifyContour(rcIntArray& points, rcIntArray& simplified,
 	{
 		int ii = (i+1) % (simplified.size()/4);
 		
-		const int ax = simplified[i*4+0];
-		const int az = simplified[i*4+2];
-		const int ai = simplified[i*4+3];
-		
-		const int bx = simplified[ii*4+0];
-		const int bz = simplified[ii*4+2];
-		const int bi = simplified[ii*4+3];
+		int ax = simplified[i*4+0];
+		int az = simplified[i*4+2];
+		int ai = simplified[i*4+3];
+
+		int bx = simplified[ii*4+0];
+		int bz = simplified[ii*4+2];
+		int bi = simplified[ii*4+3];
 
 		// Find maximum deviation from the segment.
 		float maxd = 0;
-		int i_max = -1;
+		int maxi = -1;
 		int ci, cinc, endi;
-		
+
 		// Traverse the segment in lexilogical order so that the
 		// max deviation is calculated similarly when traversing
 		// opposite segments.
@@ -338,6 +318,8 @@ static void simplifyContour(rcIntArray& points, rcIntArray& simplified,
 			cinc = pn-1;
 			ci = (bi+cinc) % pn;
 			endi = ai;
+			rcSwap(ax, bx);
+			rcSwap(az, bz);
 		}
 		
 		// Tessellate only outer edges or edges between areas.
@@ -350,7 +332,7 @@ static void simplifyContour(rcIntArray& points, rcIntArray& simplified,
 				if (d > maxd)
 				{
 					maxd = d;
-					i_max = ci;
+					maxi = ci;
 				}
 				ci = (ci+cinc) % pn;
 			}
@@ -359,7 +341,7 @@ static void simplifyContour(rcIntArray& points, rcIntArray& simplified,
 		
 		// If the max deviation is larger than accepted error,
 		// add new point, else continue to next segment.
-		if (i_max != -1 && maxd > (maxError*maxError))
+		if (maxi != -1 && maxd > (maxError*maxError))
 		{
 			// Add space for the new point.
 			simplified.resize(simplified.size()+4);
@@ -372,10 +354,10 @@ static void simplifyContour(rcIntArray& points, rcIntArray& simplified,
 				simplified[j*4+3] = simplified[(j-1)*4+3];
 			}
 			// Add the point.
-			simplified[(i+1)*4+0] = points[i_max*4+0];
-			simplified[(i+1)*4+1] = points[i_max*4+1];
-			simplified[(i+1)*4+2] = points[i_max*4+2];
-			simplified[(i+1)*4+3] = i_max;
+			simplified[(i+1)*4+0] = points[maxi*4+0];
+			simplified[(i+1)*4+1] = points[maxi*4+1];
+			simplified[(i+1)*4+2] = points[maxi*4+2];
+			simplified[(i+1)*4+3] = maxi;
 		}
 		else
 		{
@@ -397,11 +379,11 @@ static void simplifyContour(rcIntArray& points, rcIntArray& simplified,
 			const int bx = simplified[ii*4+0];
 			const int bz = simplified[ii*4+2];
 			const int bi = simplified[ii*4+3];
-
+			
 			// Find maximum deviation from the segment.
-			int i_max = -1;
+			int maxi = -1;
 			int ci = (ai+1) % pn;
-
+			
 			// Tessellate only outer edges or edges between areas.
 			bool tess = false;
 			// Wall edges.
@@ -420,22 +402,20 @@ static void simplifyContour(rcIntArray& points, rcIntArray& simplified,
 					// Round based on the segments in lexilogical order so that the
 					// max tesselation is consistent regardles in which direction
 					// segments are traversed.
-					if (bx > ax || (bx == ax && bz > az))
+					const int n = bi < ai ? (bi+pn - ai) : (bi - ai);
+					if (n > 1)
 					{
-						const int n = bi < ai ? (bi+pn - ai) : (bi - ai);
-						i_max = (ai + n/2) % pn;
-					}
-					else
-					{
-						const int n = bi < ai ? (bi+pn - ai) : (bi - ai);
-						i_max = (ai + (n+1)/2) % pn;
+						if (bx > ax || (bx == ax && bz > az))
+							maxi = (ai + n/2) % pn;
+						else
+							maxi = (ai + (n+1)/2) % pn;
 					}
 				}
 			}
 			
 			// If the max deviation is larger than accepted error,
 			// add new point, else continue to next segment.
-			if (i_max != -1)
+			if (maxi != -1)
 			{
 				// Add space for the new point.
 				simplified.resize(simplified.size()+4);
@@ -448,10 +428,10 @@ static void simplifyContour(rcIntArray& points, rcIntArray& simplified,
 					simplified[j*4+3] = simplified[(j-1)*4+3];
 				}
 				// Add the point.
-				simplified[(i+1)*4+0] = points[i_max*4+0];
-				simplified[(i+1)*4+1] = points[i_max*4+1];
-				simplified[(i+1)*4+2] = points[i_max*4+2];
-				simplified[(i+1)*4+3] = i_max;
+				simplified[(i+1)*4+0] = points[maxi*4+0];
+				simplified[(i+1)*4+1] = points[maxi*4+1];
+				simplified[(i+1)*4+2] = points[maxi*4+2];
+				simplified[(i+1)*4+3] = maxi;
 			}
 			else
 			{
@@ -466,37 +446,11 @@ static void simplifyContour(rcIntArray& points, rcIntArray& simplified,
 		// and the neighbour region is take from the next raw point.
 		const int ai = (simplified[i*4+3]+1) % pn;
 		const int bi = simplified[i*4+3];
-		simplified[i*4+3] = (points[ai*4+3] & RC_CONTOUR_REG_MASK) | (points[bi*4+3] & RC_BORDER_VERTEX);
+		simplified[i*4+3] = (points[ai*4+3] & (RC_CONTOUR_REG_MASK|RC_AREA_BORDER)) | (points[bi*4+3] & RC_BORDER_VERTEX);
 	}
 	
 }
 
-static void removeDegenerateSegments(rcIntArray& simplified)
-{
-	// Remove adjacent vertices which are equal on xz-plane,
-	// or else the triangulator will get confused.
-	for (int i = 0; i < simplified.size()/4; ++i)
-	{
-		int ni = i+1;
-		if (ni >= (simplified.size()/4))
-			ni = 0;
-			
-		if (simplified[i*4+0] == simplified[ni*4+0] &&
-			simplified[i*4+2] == simplified[ni*4+2])
-		{
-			// Degenerate segment, remove.
-			for (int j = i; j < simplified.size()/4-1; ++j)
-			{
-				simplified[j*4+0] = simplified[(j+1)*4+0];
-				simplified[j*4+1] = simplified[(j+1)*4+1];
-				simplified[j*4+2] = simplified[(j+1)*4+2];
-				simplified[j*4+3] = simplified[(j+1)*4+3];
-			}
-			simplified.resize(simplified.size()-4);
-		}
-	}
-}
-
 static int calcAreaOfPolygon2D(const int* verts, const int nverts)
 {
 	int area = 0;
@@ -509,54 +463,155 @@ static int calcAreaOfPolygon2D(const int* verts, const int nverts)
 	return (area+1) / 2;
 }
 
-inline bool ileft(const int* a, const int* b, const int* c)
+// TODO: these are the same as in RecastMesh.cpp, consider using the same.
+// Last time I checked the if version got compiled using cmov, which was a lot faster than module (with idiv).
+inline int prev(int i, int n) { return i-1 >= 0 ? i-1 : n-1; }
+inline int next(int i, int n) { return i+1 < n ? i+1 : 0; }
+
+inline int area2(const int* a, const int* b, const int* c)
 {
-	return (b[0] - a[0]) * (c[2] - a[2]) - (c[0] - a[0]) * (b[2] - a[2]) <= 0;
+	return (b[0] - a[0]) * (c[2] - a[2]) - (c[0] - a[0]) * (b[2] - a[2]);
 }
 
-static void getClosestIndices(const int* vertsa, const int nvertsa,
-							  const int* vertsb, const int nvertsb,
-							  int& ia, int& ib)
+//	Exclusive or: true iff exactly one argument is true.
+//	The arguments are negated to ensure that they are 0/1
+//	values.  Then the bitwise Xor operator may apply.
+//	(This idea is due to Michael Baldwin.)
+inline bool xorb(bool x, bool y)
 {
-	int closestDist = 0xfffffff;
-	ia = -1, ib = -1;
-	for (int i = 0; i < nvertsa; ++i)
+	return !x ^ !y;
+}
+
+// Returns true iff c is strictly to the left of the directed
+// line through a to b.
+inline bool left(const int* a, const int* b, const int* c)
+{
+	return area2(a, b, c) < 0;
+}
+
+inline bool leftOn(const int* a, const int* b, const int* c)
+{
+	return area2(a, b, c) <= 0;
+}
+
+inline bool collinear(const int* a, const int* b, const int* c)
+{
+	return area2(a, b, c) == 0;
+}
+
+//	Returns true iff ab properly intersects cd: they share
+//	a point interior to both segments.  The properness of the
+//	intersection is ensured by using strict leftness.
+static bool intersectProp(const int* a, const int* b, const int* c, const int* d)
+{
+	// Eliminate improper cases.
+	if (collinear(a,b,c) || collinear(a,b,d) ||
+		collinear(c,d,a) || collinear(c,d,b))
+		return false;
+	
+	return xorb(left(a,b,c), left(a,b,d)) && xorb(left(c,d,a), left(c,d,b));
+}
+
+// Returns T iff (a,b,c) are collinear and point c lies
+// on the closed segement ab.
+static bool between(const int* a, const int* b, const int* c)
+{
+	if (!collinear(a, b, c))
+		return false;
+	// If ab not vertical, check betweenness on x; else on y.
+	if (a[0] != b[0])
+		return	((a[0] <= c[0]) && (c[0] <= b[0])) || ((a[0] >= c[0]) && (c[0] >= b[0]));
+	else
+		return	((a[2] <= c[2]) && (c[2] <= b[2])) || ((a[2] >= c[2]) && (c[2] >= b[2]));
+}
+
+// Returns true iff segments ab and cd intersect, properly or improperly.
+static bool intersect(const int* a, const int* b, const int* c, const int* d)
+{
+	if (intersectProp(a, b, c, d))
+		return true;
+	else if (between(a, b, c) || between(a, b, d) ||
+			 between(c, d, a) || between(c, d, b))
+		return true;
+	else
+		return false;
+}
+
+static bool vequal(const int* a, const int* b)
+{
+	return a[0] == b[0] && a[2] == b[2];
+}
+
+static bool intersectSegCountour(const int* d0, const int* d1, int i, int n, const int* verts)
+{
+	// For each edge (k,k+1) of P
+	for (int k = 0; k < n; k++)
 	{
-		const int in = (i+1) % nvertsa;
-		const int ip = (i+nvertsa-1) % nvertsa;
-		const int* va = &vertsa[i*4];
-		const int* van = &vertsa[in*4];
-		const int* vap = &vertsa[ip*4];
+		int k1 = next(k, n);
+		// Skip edges incident to i.
+		if (i == k || i == k1)
+			continue;
+		const int* p0 = &verts[k * 4];
+		const int* p1 = &verts[k1 * 4];
+		if (vequal(d0, p0) || vequal(d1, p0) || vequal(d0, p1) || vequal(d1, p1))
+			continue;
 		
-		for (int j = 0; j < nvertsb; ++j)
+		if (intersect(d0, d1, p0, p1))
+			return true;
+	}
+	return false;
+}
+
+static bool	inCone(int i, int n, const int* verts, const int* pj)
+{
+	const int* pi = &verts[i * 4];
+	const int* pi1 = &verts[next(i, n) * 4];
+	const int* pin1 = &verts[prev(i, n) * 4];
+	
+	// If P[i] is a convex vertex [ i+1 left or on (i-1,i) ].
+	if (leftOn(pin1, pi, pi1))
+		return left(pi, pj, pin1) && left(pj, pi, pi1);
+	// Assume (i-1,i,i+1) not collinear.
+	// else P[i] is reflex.
+	return !(leftOn(pi, pj, pi1) && leftOn(pj, pi, pin1));
+}
+
+
+static void removeDegenerateSegments(rcIntArray& simplified)
+{
+	// Remove adjacent vertices which are equal on xz-plane,
+	// or else the triangulator will get confused.
+	int npts = simplified.size()/4;
+	for (int i = 0; i < npts; ++i)
+	{
+		int ni = next(i, npts);
+		
+		if (vequal(&simplified[i*4], &simplified[ni*4]))
 		{
-			const int* vb = &vertsb[j*4];
-			// vb must be "infront" of va.
-			if (ileft(vap,va,vb) && ileft(va,van,vb))
+			// Degenerate segment, remove.
+			for (int j = i; j < simplified.size()/4-1; ++j)
 			{
-				const int dx = vb[0] - va[0];
-				const int dz = vb[2] - va[2];
-				const int d = dx*dx + dz*dz;
-				if (d < closestDist)
-				{
-					ia = i;
-					ib = j;
-					closestDist = d;
-				}
+				simplified[j*4+0] = simplified[(j+1)*4+0];
+				simplified[j*4+1] = simplified[(j+1)*4+1];
+				simplified[j*4+2] = simplified[(j+1)*4+2];
+				simplified[j*4+3] = simplified[(j+1)*4+3];
 			}
+			simplified.resize(simplified.size()-4);
+			npts--;
 		}
 	}
 }
 
+
 static bool mergeContours(rcContour& ca, rcContour& cb, int ia, int ib)
 {
 	const int maxVerts = ca.nverts + cb.nverts + 2;
 	int* verts = (int*)rcAlloc(sizeof(int)*maxVerts*4, RC_ALLOC_PERM);
 	if (!verts)
 		return false;
-
+	
 	int nv = 0;
-
+	
 	// Copy contour A.
 	for (int i = 0; i <= ca.nverts; ++i)
 	{
@@ -584,7 +639,7 @@ static bool mergeContours(rcContour& ca, rcContour& cb, int ia, int ib)
 	rcFree(ca.verts);
 	ca.verts = verts;
 	ca.nverts = nv;
-
+	
 	rcFree(cb.verts);
 	cb.verts = 0;
 	cb.nverts = 0;
@@ -592,18 +647,179 @@ static bool mergeContours(rcContour& ca, rcContour& cb, int ia, int ib)
 	return true;
 }
 
+struct rcContourHole
+{
+	rcContour* contour;
+	int minx, minz, leftmost;
+};
+
+struct rcContourRegion
+{
+	rcContour* outline;
+	rcContourHole* holes;
+	int nholes;
+};
+
+struct rcPotentialDiagonal
+{
+	int vert;
+	int dist;
+};
+
+// Finds the lowest leftmost vertex of a contour.
+static void findLeftMostVertex(rcContour* contour, int* minx, int* minz, int* leftmost)
+{
+	*minx = contour->verts[0];
+	*minz = contour->verts[2];
+	*leftmost = 0;
+	for (int i = 1; i < contour->nverts; i++)
+	{
+		const int x = contour->verts[i*4+0];
+		const int z = contour->verts[i*4+2];
+		if (x < *minx || (x == *minx && z < *minz))
+		{
+			*minx = x;
+			*minz = z;
+			*leftmost = i;
+		}
+	}
+}
+
+static int compareHoles(const void* va, const void* vb)
+{
+	const rcContourHole* a = (const rcContourHole*)va;
+	const rcContourHole* b = (const rcContourHole*)vb;
+	if (a->minx == b->minx)
+	{
+		if (a->minz < b->minz)
+			return -1;
+		if (a->minz > b->minz)
+			return 1;
+	}
+	else
+	{
+		if (a->minx < b->minx)
+			return -1;
+		if (a->minx > b->minx)
+			return 1;
+	}
+	return 0;
+}
+
+
+static int compareDiagDist(const void* va, const void* vb)
+{
+	const rcPotentialDiagonal* a = (const rcPotentialDiagonal*)va;
+	const rcPotentialDiagonal* b = (const rcPotentialDiagonal*)vb;
+	if (a->dist < b->dist)
+		return -1;
+	if (a->dist > b->dist)
+		return 1;
+	return 0;
+}
+
+
+static void mergeRegionHoles(rcContext* ctx, rcContourRegion& region)
+{
+	// Sort holes from left to right.
+	for (int i = 0; i < region.nholes; i++)
+		findLeftMostVertex(region.holes[i].contour, &region.holes[i].minx, &region.holes[i].minz, &region.holes[i].leftmost);
+	
+	qsort(region.holes, region.nholes, sizeof(rcContourHole), compareHoles);
+	
+	int maxVerts = region.outline->nverts;
+	for (int i = 0; i < region.nholes; i++)
+		maxVerts += region.holes[i].contour->nverts;
+	
+	rcScopedDelete<rcPotentialDiagonal> diags((rcPotentialDiagonal*)rcAlloc(sizeof(rcPotentialDiagonal)*maxVerts, RC_ALLOC_TEMP));
+	if (!diags)
+	{
+		ctx->log(RC_LOG_WARNING, "mergeRegionHoles: Failed to allocated diags %d.", maxVerts);
+		return;
+	}
+	
+	rcContour* outline = region.outline;
+	
+	// Merge holes into the outline one by one.
+	for (int i = 0; i < region.nholes; i++)
+	{
+		rcContour* hole = region.holes[i].contour;
+		
+		int index = -1;
+		int bestVertex = region.holes[i].leftmost;
+		for (int iter = 0; iter < hole->nverts; iter++)
+		{
+			// Find potential diagonals.
+			// The 'best' vertex must be in the cone described by 3 cosequtive vertices of the outline.
+			// ..o j-1
+			//   |
+			//   |   * best
+			//   |
+			// j o-----o j+1
+			//         :
+			int ndiags = 0;
+			const int* corner = &hole->verts[bestVertex*4];
+			for (int j = 0; j < outline->nverts; j++)
+			{
+				if (inCone(j, outline->nverts, outline->verts, corner))
+				{
+					int dx = outline->verts[j*4+0] - corner[0];
+					int dz = outline->verts[j*4+2] - corner[2];
+					diags[ndiags].vert = j;
+					diags[ndiags].dist = dx*dx + dz*dz;
+					ndiags++;
+				}
+			}
+			// Sort potential diagonals by distance, we want to make the connection as short as possible.
+			qsort(diags, ndiags, sizeof(rcPotentialDiagonal), compareDiagDist);
+			
+			// Find a diagonal that is not intersecting the outline not the remaining holes.
+			index = -1;
+			for (int j = 0; j < ndiags; j++)
+			{
+				const int* pt = &outline->verts[diags[j].vert*4];
+				bool intersect = intersectSegCountour(pt, corner, diags[i].vert, outline->nverts, outline->verts);
+				for (int k = i; k < region.nholes && !intersect; k++)
+					intersect |= intersectSegCountour(pt, corner, -1, region.holes[k].contour->nverts, region.holes[k].contour->verts);
+				if (!intersect)
+				{
+					index = diags[j].vert;
+					break;
+				}
+			}
+			// If found non-intersecting diagonal, stop looking.
+			if (index != -1)
+				break;
+			// All the potential diagonals for the current vertex were intersecting, try next vertex.
+			bestVertex = (bestVertex + 1) % hole->nverts;
+		}
+		
+		if (index == -1)
+		{
+			ctx->log(RC_LOG_WARNING, "mergeHoles: Failed to find merge points for %p and %p.", region.outline, hole);
+			continue;
+		}
+		if (!mergeContours(*region.outline, *hole, index, bestVertex))
+		{
+			ctx->log(RC_LOG_WARNING, "mergeHoles: Failed to merge contours %p and %p.", region.outline, hole);
+			continue;
+		}
+	}
+}
+
+
 /// @par
 ///
 /// The raw contours will match the region outlines exactly. The @p maxError and @p maxEdgeLen
 /// parameters control how closely the simplified contours will match the raw contours.
 ///
-/// Simplified contours are generated such that the vertices for portals between areas match up. 
+/// Simplified contours are generated such that the vertices for portals between areas match up.
 /// (They are considered mandatory vertices.)
 ///
 /// Setting @p maxEdgeLength to zero will disabled the edge length feature.
-/// 
+///
 /// See the #rcConfig documentation for more information on the configuration parameters.
-/// 
+///
 /// @see rcAllocContourSet, rcCompactHeightfield, rcContourSet, rcConfig
 bool rcBuildContours(rcContext* ctx, rcCompactHeightfield& chf,
 					 const float maxError, const int maxEdgeLen,
@@ -615,7 +831,7 @@ bool rcBuildContours(rcContext* ctx, rcCompactHeightfield& chf,
 	const int h = chf.height;
 	const int borderSize = chf.borderSize;
 	
-	ctx->startTimer(RC_TIMER_BUILD_CONTOURS);
+	rcScopedTimer timer(ctx, RC_TIMER_BUILD_CONTOURS);
 	
 	rcVcopy(cset.bmin, chf.bmin);
 	rcVcopy(cset.bmax, chf.bmax);
@@ -633,6 +849,7 @@ bool rcBuildContours(rcContext* ctx, rcCompactHeightfield& chf,
 	cset.width = chf.width - chf.borderSize*2;
 	cset.height = chf.height - chf.borderSize*2;
 	cset.borderSize = chf.borderSize;
+	cset.maxError = maxError;
 	
 	int maxContours = rcMax((int)chf.maxRegions, 8);
 	cset.conts = (rcContour*)rcAlloc(sizeof(rcContour)*maxContours, RC_ALLOC_PERM);
@@ -640,7 +857,7 @@ bool rcBuildContours(rcContext* ctx, rcCompactHeightfield& chf,
 		return false;
 	cset.nconts = 0;
 	
-	rcScopedDelete<unsigned char> flags = (unsigned char*)rcAlloc(sizeof(unsigned char)*chf.spanCount, RC_ALLOC_TEMP);
+	rcScopedDelete<unsigned char> flags((unsigned char*)rcAlloc(sizeof(unsigned char)*chf.spanCount, RC_ALLOC_TEMP));
 	if (!flags)
 	{
 		ctx->log(RC_LOG_ERROR, "rcBuildContours: Out of memory 'flags' (%d).", chf.spanCount);
@@ -706,17 +923,17 @@ bool rcBuildContours(rcContext* ctx, rcCompactHeightfield& chf,
 				
 				verts.resize(0);
 				simplified.resize(0);
-
+				
 				ctx->startTimer(RC_TIMER_BUILD_CONTOURS_TRACE);
 				walkContour(x, y, i, chf, flags, verts);
 				ctx->stopTimer(RC_TIMER_BUILD_CONTOURS_TRACE);
-
+				
 				ctx->startTimer(RC_TIMER_BUILD_CONTOURS_SIMPLIFY);
 				simplifyContour(verts, simplified, maxError, maxEdgeLen, buildFlags);
 				removeDegenerateSegments(simplified);
 				ctx->stopTimer(RC_TIMER_BUILD_CONTOURS_SIMPLIFY);
 				
-
+				
 				// Store region->contour remap info.
 				// Create contour.
 				if (simplified.size()/4 >= 3)
@@ -724,7 +941,7 @@ bool rcBuildContours(rcContext* ctx, rcCompactHeightfield& chf,
 					if (cset.nconts >= maxContours)
 					{
 						// Allocate more contours.
-						// This can happen when there are tiny holes in the heightfield.
+						// This happens when a region has holes.
 						const int oldMax = maxContours;
 						maxContours *= 2;
 						rcContour* newConts = (rcContour*)rcAlloc(sizeof(rcContour)*maxContours, RC_ALLOC_PERM);
@@ -737,10 +954,10 @@ bool rcBuildContours(rcContext* ctx, rcCompactHeightfield& chf,
 						}
 						rcFree(cset.conts);
 						cset.conts = newConts;
-					
+						
 						ctx->log(RC_LOG_WARNING, "rcBuildContours: Expanding max contours from %d to %d.", oldMax, maxContours);
 					}
-						
+					
 					rcContour* cont = &cset.conts[cset.nconts++];
 					
 					cont->nverts = simplified.size()/4;
@@ -754,9 +971,9 @@ bool rcBuildContours(rcContext* ctx, rcCompactHeightfield& chf,
 					if (borderSize > 0)
 					{
 						// If the heightfield was build with bordersize, remove the offset.
-						for (int i = 0; i < cont->nverts; ++i)
+						for (int j = 0; j < cont->nverts; ++j)
 						{
-							int* v = &cont->verts[i*4];
+							int* v = &cont->verts[j*4];
 							v[0] -= borderSize;
 							v[2] -= borderSize;
 						}
@@ -773,25 +990,14 @@ bool rcBuildContours(rcContext* ctx, rcCompactHeightfield& chf,
 					if (borderSize > 0)
 					{
 						// If the heightfield was build with bordersize, remove the offset.
-						for (int i = 0; i < cont->nrverts; ++i)
+						for (int j = 0; j < cont->nrverts; ++j)
 						{
-							int* v = &cont->rverts[i*4];
+							int* v = &cont->rverts[j*4];
 							v[0] -= borderSize;
 							v[2] -= borderSize;
 						}
 					}
 					
-/*					cont->cx = cont->cy = cont->cz = 0;
-					for (int i = 0; i < cont->nverts; ++i)
-					{
-						cont->cx += cont->verts[i*4+0];
-						cont->cy += cont->verts[i*4+1];
-						cont->cz += cont->verts[i*4+2];
-					}
-					cont->cx /= cont->nverts;
-					cont->cy /= cont->nverts;
-					cont->cz /= cont->nverts;*/
-					
 					cont->reg = reg;
 					cont->area = area;
 				}
@@ -799,55 +1005,101 @@ bool rcBuildContours(rcContext* ctx, rcCompactHeightfield& chf,
 		}
 	}
 	
-	// Check and merge droppings.
-	// Sometimes the previous algorithms can fail and create several contours
-	// per area. This pass will try to merge the holes into the main region.
-	for (int i = 0; i < cset.nconts; ++i)
+	// Merge holes if needed.
+	if (cset.nconts > 0)
 	{
-		rcContour& cont = cset.conts[i];
-		// Check if the contour is would backwards.
-		if (calcAreaOfPolygon2D(cont.verts, cont.nverts) < 0)
+		// Calculate winding of all polygons.
+		rcScopedDelete<char> winding((char*)rcAlloc(sizeof(char)*cset.nconts, RC_ALLOC_TEMP));
+		if (!winding)
 		{
-			// Find another contour which has the same region ID.
-			int mergeIdx = -1;
-			for (int j = 0; j < cset.nconts; ++j)
+			ctx->log(RC_LOG_ERROR, "rcBuildContours: Out of memory 'hole' (%d).", cset.nconts);
+			return false;
+		}
+		int nholes = 0;
+		for (int i = 0; i < cset.nconts; ++i)
+		{
+			rcContour& cont = cset.conts[i];
+			// If the contour is wound backwards, it is a hole.
+			winding[i] = calcAreaOfPolygon2D(cont.verts, cont.nverts) < 0 ? -1 : 1;
+			if (winding[i] < 0)
+				nholes++;
+		}
+		
+		if (nholes > 0)
+		{
+			// Collect outline contour and holes contours per region.
+			// We assume that there is one outline and multiple holes.
+			const int nregions = chf.maxRegions+1;
+			rcScopedDelete<rcContourRegion> regions((rcContourRegion*)rcAlloc(sizeof(rcContourRegion)*nregions, RC_ALLOC_TEMP));
+			if (!regions)
 			{
-				if (i == j) continue;
-				if (cset.conts[j].nverts && cset.conts[j].reg == cont.reg)
+				ctx->log(RC_LOG_ERROR, "rcBuildContours: Out of memory 'regions' (%d).", nregions);
+				return false;
+			}
+			memset(regions, 0, sizeof(rcContourRegion)*nregions);
+			
+			rcScopedDelete<rcContourHole> holes((rcContourHole*)rcAlloc(sizeof(rcContourHole)*cset.nconts, RC_ALLOC_TEMP));
+			if (!holes)
+			{
+				ctx->log(RC_LOG_ERROR, "rcBuildContours: Out of memory 'holes' (%d).", cset.nconts);
+				return false;
+			}
+			memset(holes, 0, sizeof(rcContourHole)*cset.nconts);
+			
+			for (int i = 0; i < cset.nconts; ++i)
+			{
+				rcContour& cont = cset.conts[i];
+				// Positively would contours are outlines, negative holes.
+				if (winding[i] > 0)
 				{
-					// Make sure the polygon is correctly oriented.
-					if (calcAreaOfPolygon2D(cset.conts[j].verts, cset.conts[j].nverts))
-					{
-						mergeIdx = j;
-						break;
-					}
+					if (regions[cont.reg].outline)
+						ctx->log(RC_LOG_ERROR, "rcBuildContours: Multiple outlines for region %d.", cont.reg);
+					regions[cont.reg].outline = &cont;
+				}
+				else
+				{
+					regions[cont.reg].nholes++;
 				}
 			}
-			if (mergeIdx == -1)
+			int index = 0;
+			for (int i = 0; i < nregions; i++)
 			{
-				ctx->log(RC_LOG_WARNING, "rcBuildContours: Could not find merge target for bad contour %d.", i);
-			}
-			else
-			{
-				rcContour& mcont = cset.conts[mergeIdx];
-				// Merge by closest points.
-				int ia = 0, ib = 0;
-				getClosestIndices(mcont.verts, mcont.nverts, cont.verts, cont.nverts, ia, ib);
-				if (ia == -1 || ib == -1)
+				if (regions[i].nholes > 0)
 				{
-					ctx->log(RC_LOG_WARNING, "rcBuildContours: Failed to find merge points for %d and %d.", i, mergeIdx);
-					continue;
+					regions[i].holes = &holes[index];
+					index += regions[i].nholes;
+					regions[i].nholes = 0;
 				}
-				if (!mergeContours(mcont, cont, ia, ib))
+			}
+			for (int i = 0; i < cset.nconts; ++i)
+			{
+				rcContour& cont = cset.conts[i];
+				rcContourRegion& reg = regions[cont.reg];
+				if (winding[i] < 0)
+					reg.holes[reg.nholes++].contour = &cont;
+			}
+			
+			// Finally merge each regions holes into the outline.
+			for (int i = 0; i < nregions; i++)
+			{
+				rcContourRegion& reg = regions[i];
+				if (!reg.nholes) continue;
+				
+				if (reg.outline)
 				{
-					ctx->log(RC_LOG_WARNING, "rcBuildContours: Failed to merge contours %d and %d.", i, mergeIdx);
-					continue;
+					mergeRegionHoles(ctx, reg);
+				}
+				else
+				{
+					// The region does not have an outline.
+					// This can happen if the contour becaomes selfoverlapping because of
+					// too aggressive simplification settings.
+					ctx->log(RC_LOG_ERROR, "rcBuildContours: Bad outline for region %d, contour simplification is likely too aggressive.", i);
 				}
 			}
 		}
+		
 	}
 	
-	ctx->stopTimer(RC_TIMER_BUILD_CONTOURS);
-	
 	return true;
 }

extern/recastnavigation/Recast/Source/RecastFilter.cpp

@@ -37,7 +37,7 @@ void rcFilterLowHangingWalkableObstacles(rcContext* ctx, const int walkableClimb
 {
 	rcAssert(ctx);
 
-	ctx->startTimer(RC_TIMER_FILTER_LOW_OBSTACLES);
+	rcScopedTimer timer(ctx, RC_TIMER_FILTER_LOW_OBSTACLES);
 	
 	const int w = solid.width;
 	const int h = solid.height;
@@ -67,8 +67,6 @@ void rcFilterLowHangingWalkableObstacles(rcContext* ctx, const int walkableClimb
 			}
 		}
 	}
-
-	ctx->stopTimer(RC_TIMER_FILTER_LOW_OBSTACLES);
 }
 
 /// @par
@@ -86,7 +84,7 @@ void rcFilterLedgeSpans(rcContext* ctx, const int walkableHeight, const int walk
 {
 	rcAssert(ctx);
 	
-	ctx->startTimer(RC_TIMER_FILTER_BORDER);
+	rcScopedTimer timer(ctx, RC_TIMER_FILTER_BORDER);
 
 	const int w = solid.width;
 	const int h = solid.height;
@@ -128,7 +126,7 @@ void rcFilterLedgeSpans(rcContext* ctx, const int walkableHeight, const int walk
 					rcSpan* ns = solid.spans[dx + dy*w];
 					int nbot = -walkableClimb;
 					int ntop = ns ? (int)ns->smin : MAX_HEIGHT;
-					// Skip neighbor if the gap between the spans is too small.
+					// Skip neightbour if the gap between the spans is too small.
 					if (rcMin(top,ntop) - rcMax(bot,nbot) > walkableHeight)
 						minh = rcMin(minh, nbot - bot);
 					
@@ -137,7 +135,7 @@ void rcFilterLedgeSpans(rcContext* ctx, const int walkableHeight, const int walk
 					{
 						nbot = (int)ns->smax;
 						ntop = ns->next ? (int)ns->next->smin : MAX_HEIGHT;
-						// Skip neighbor if the gap between the spans is too small.
+						// Skip neightbour if the gap between the spans is too small.
 						if (rcMin(top,ntop) - rcMax(bot,nbot) > walkableHeight)
 						{
 							minh = rcMin(minh, nbot - bot);
@@ -156,20 +154,19 @@ void rcFilterLedgeSpans(rcContext* ctx, const int walkableHeight, const int walk
 				// The current span is close to a ledge if the drop to any
 				// neighbour span is less than the walkableClimb.
 				if (minh < -walkableClimb)
+				{
 					s->area = RC_NULL_AREA;
-					
+				}
 				// If the difference between all neighbours is too large,
 				// we are at steep slope, mark the span as ledge.
-				if ((asmax - asmin) > walkableClimb)
+				else if ((asmax - asmin) > walkableClimb)
 				{
 					s->area = RC_NULL_AREA;
 				}
 			}
 		}
 	}
-	
-	ctx->stopTimer(RC_TIMER_FILTER_BORDER);
-}	
+}
 
 /// @par
 ///
@@ -181,7 +178,7 @@ void rcFilterWalkableLowHeightSpans(rcContext* ctx, int walkableHeight, rcHeight
 {
 	rcAssert(ctx);
 	
-	ctx->startTimer(RC_TIMER_FILTER_WALKABLE);
+	rcScopedTimer timer(ctx, RC_TIMER_FILTER_WALKABLE);
 	
 	const int w = solid.width;
 	const int h = solid.height;
@@ -202,6 +199,4 @@ void rcFilterWalkableLowHeightSpans(rcContext* ctx, int walkableHeight, rcHeight
 			}
 		}
 	}
-	
-	ctx->stopTimer(RC_TIMER_FILTER_WALKABLE);
 }

extern/recastnavigation/Recast/Source/RecastLayers.cpp

@@ -38,7 +38,7 @@ struct rcLayerRegion
 	unsigned char layerId;		// Layer ID
 	unsigned char nlayers;		// Layer count
 	unsigned char nneis;		// Neighbour count
-	unsigned char base;			// Flag indicating if the region is hte base of merged regions.
+	unsigned char base;		// Flag indicating if the region is the base of merged regions.
 };
 
 
@@ -87,12 +87,12 @@ bool rcBuildHeightfieldLayers(rcContext* ctx, rcCompactHeightfield& chf,
 {
 	rcAssert(ctx);
 	
-	ctx->startTimer(RC_TIMER_BUILD_LAYERS);
+	rcScopedTimer timer(ctx, RC_TIMER_BUILD_LAYERS);
 	
 	const int w = chf.width;
 	const int h = chf.height;
 	
-	rcScopedDelete<unsigned char> srcReg = (unsigned char*)rcAlloc(sizeof(unsigned char)*chf.spanCount, RC_ALLOC_TEMP);
+	rcScopedDelete<unsigned char> srcReg((unsigned char*)rcAlloc(sizeof(unsigned char)*chf.spanCount, RC_ALLOC_TEMP));
 	if (!srcReg)
 	{
 		ctx->log(RC_LOG_ERROR, "rcBuildHeightfieldLayers: Out of memory 'srcReg' (%d).", chf.spanCount);
@@ -101,7 +101,7 @@ bool rcBuildHeightfieldLayers(rcContext* ctx, rcCompactHeightfield& chf,
 	memset(srcReg,0xff,sizeof(unsigned char)*chf.spanCount);
 	
 	const int nsweeps = chf.width;
-	rcScopedDelete<rcLayerSweepSpan> sweeps = (rcLayerSweepSpan*)rcAlloc(sizeof(rcLayerSweepSpan)*nsweeps, RC_ALLOC_TEMP);
+	rcScopedDelete<rcLayerSweepSpan> sweeps((rcLayerSweepSpan*)rcAlloc(sizeof(rcLayerSweepSpan)*nsweeps, RC_ALLOC_TEMP));
 	if (!sweeps)
 	{
 		ctx->log(RC_LOG_ERROR, "rcBuildHeightfieldLayers: Out of memory 'sweeps' (%d).", nsweeps);
@@ -212,7 +212,7 @@ bool rcBuildHeightfieldLayers(rcContext* ctx, rcCompactHeightfield& chf,
 
 	// Allocate and init layer regions.
 	const int nregs = (int)regId;
-	rcScopedDelete<rcLayerRegion> regs = (rcLayerRegion*)rcAlloc(sizeof(rcLayerRegion)*nregs, RC_ALLOC_TEMP);
+	rcScopedDelete<rcLayerRegion> regs((rcLayerRegion*)rcAlloc(sizeof(rcLayerRegion)*nregs, RC_ALLOC_TEMP));
 	if (!regs)
 	{
 		ctx->log(RC_LOG_ERROR, "rcBuildHeightfieldLayers: Out of memory 'regs' (%d).", nregs);
@@ -258,7 +258,7 @@ bool rcBuildHeightfieldLayers(rcContext* ctx, rcCompactHeightfield& chf,
 						const int ay = y + rcGetDirOffsetY(dir);
 						const int ai = (int)chf.cells[ax+ay*w].index + rcGetCon(s, dir);
 						const unsigned char rai = srcReg[ai];
-						if (rai != 0xff && rai != ri)
+						if (rai != 0xff && rai != ri && regs[ri].nneis < RC_MAX_NEIS)
 							addUnique(regs[ri].neis, regs[ri].nneis, rai);
 					}
 				}
@@ -293,7 +293,7 @@ bool rcBuildHeightfieldLayers(rcContext* ctx, rcCompactHeightfield& chf,
 	for (int i = 0; i < nregs; ++i)
 	{
 		rcLayerRegion& root = regs[i];
-		// Skip alreadu visited.
+		// Skip already visited.
 		if (root.layerId != 0xff)
 			continue;
 
@@ -325,7 +325,7 @@ bool rcBuildHeightfieldLayers(rcContext* ctx, rcCompactHeightfield& chf,
 					continue;
 				// Skip if the height range would become too large.
 				const int ymin = rcMin(root.ymin, regn.ymin);
-				const int ymax = rcMin(root.ymax, regn.ymax);
+				const int ymax = rcMax(root.ymax, regn.ymax);
 				if ((ymax - ymin) >= 255)
 					 continue;
 
@@ -373,11 +373,11 @@ bool rcBuildHeightfieldLayers(rcContext* ctx, rcCompactHeightfield& chf,
 					continue;
 				// Skip if the height range would become too large.
 				const int ymin = rcMin(ri.ymin, rj.ymin);
-				const int ymax = rcMin(ri.ymax, rj.ymax);
+				const int ymax = rcMax(ri.ymax, rj.ymax);
 				if ((ymax - ymin) >= 255)
 				  continue;
 						  
-				// Make sure that there is no overlap when mergin 'ri' and 'rj'.
+				// Make sure that there is no overlap when merging 'ri' and 'rj'.
 				bool overlap = false;
 				// Iterate over all regions which have the same layerId as 'rj'
 				for (int k = 0; k < nregs; ++k)
@@ -417,7 +417,7 @@ bool rcBuildHeightfieldLayers(rcContext* ctx, rcCompactHeightfield& chf,
 					// Add overlaid layers from 'rj' to 'ri'.
 					for (int k = 0; k < rj.nlayers; ++k)
 						addUnique(ri.layers, ri.nlayers, rj.layers[k]);
-					// Update heigh bounds.
+					// Update height bounds.
 					ri.ymin = rcMin(ri.ymin, rj.ymin);
 					ri.ymax = rcMax(ri.ymax, rj.ymax);
 				}
@@ -446,10 +446,7 @@ bool rcBuildHeightfieldLayers(rcContext* ctx, rcCompactHeightfield& chf,
 	
 	// No layers, return empty.
 	if (layerId == 0)
-	{
-		ctx->stopTimer(RC_TIMER_BUILD_LAYERS);
 		return true;
-	}
 	
 	// Create layers.
 	rcAssert(lset.layers == 0);
@@ -481,10 +478,8 @@ bool rcBuildHeightfieldLayers(rcContext* ctx, rcCompactHeightfield& chf,
 	for (int i = 0; i < lset.nlayers; ++i)
 	{
 		unsigned char curId = (unsigned char)i;
-		
-		// Allocate memory for the current layer.
+
 		rcHeightfieldLayer* layer = &lset.layers[i];
-		memset(layer, 0, sizeof(rcHeightfieldLayer));
 
 		const int gridSize = sizeof(unsigned char)*lw*lh;
 
@@ -528,7 +523,7 @@ bool rcBuildHeightfieldLayers(rcContext* ctx, rcCompactHeightfield& chf,
 		layer->cs = chf.cs;
 		layer->ch = chf.ch;
 		
-		// Adjust the bbox to fit the heighfield.
+		// Adjust the bbox to fit the heightfield.
 		rcVcopy(layer->bmin, bmin);
 		rcVcopy(layer->bmax, bmax);
 		layer->bmin[1] = bmin[1] + hmin*chf.ch;
@@ -542,7 +537,7 @@ bool rcBuildHeightfieldLayers(rcContext* ctx, rcCompactHeightfield& chf,
 		layer->miny = layer->height;
 		layer->maxy = 0;
 		
-		// Copy height and area from compact heighfield. 
+		// Copy height and area from compact heightfield. 
 		for (int y = 0; y < lh; ++y)
 		{
 			for (int x = 0; x < lw; ++x)
@@ -550,14 +545,14 @@ bool rcBuildHeightfieldLayers(rcContext* ctx, rcCompactHeightfield& chf,
 				const int cx = borderSize+x;
 				const int cy = borderSize+y;
 				const rcCompactCell& c = chf.cells[cx+cy*w];
-				for (int i = (int)c.index, ni = (int)(c.index+c.count); i < ni; ++i)
+				for (int j = (int)c.index, nj = (int)(c.index+c.count); j < nj; ++j)
 				{
-					const rcCompactSpan& s = chf.spans[i];
+					const rcCompactSpan& s = chf.spans[j];
 					// Skip unassigned regions.
-					if (srcReg[i] == 0xff)
+					if (srcReg[j] == 0xff)
 						continue;
 					// Skip of does nto belong to current layer.
-					unsigned char lid = regs[srcReg[i]].layerId;
+					unsigned char lid = regs[srcReg[j]].layerId;
 					if (lid != curId)
 						continue;
 					
@@ -570,7 +565,7 @@ bool rcBuildHeightfieldLayers(rcContext* ctx, rcCompactHeightfield& chf,
 					// Store height and area type.
 					const int idx = x+y*lw;
 					layer->heights[idx] = (unsigned char)(s.y - hmin);
-					layer->areas[idx] = chf.areas[i];
+					layer->areas[idx] = chf.areas[j];
 					
 					// Check connection.
 					unsigned char portal = 0;
@@ -614,7 +609,5 @@ bool rcBuildHeightfieldLayers(rcContext* ctx, rcCompactHeightfield& chf,
 			layer->miny = layer->maxy = 0;
 	}
 	
-	ctx->stopTimer(RC_TIMER_BUILD_LAYERS);
-	
 	return true;
 }

extern/recastnavigation/Recast/Source/RecastLog.cpp

@@ -1,77 +0,0 @@
-//
-// Copyright (c) 2009 Mikko Mononen memon@inside.org
-//
-// This software is provided 'as-is', without any express or implied
-// warranty.  In no event will the authors be held liable for any damages
-// arising from the use of this software.
-// Permission is granted to anyone to use this software for any purpose,
-// including commercial applications, and to alter it and redistribute it
-// freely, subject to the following restrictions:
-// 1. The origin of this software must not be misrepresented; you must not
-//    claim that you wrote the original software. If you use this software
-//    in a product, an acknowledgment in the product documentation would be
-//    appreciated but is not required.
-// 2. Altered source versions must be plainly marked as such, and must not be
-//    misrepresented as being the original software.
-// 3. This notice may not be removed or altered from any source distribution.
-//
-
-#include "RecastLog.h"
-#include <stdio.h>
-#include <stdarg.h>
-
-static rcLog* g_log = 0;
-static rcBuildTimes* g_btimes = 0;
-
-rcLog::rcLog() :
-	m_messageCount(0),
-	m_textPoolSize(0)
-{
-}
-
-rcLog::~rcLog()
-{
-	if (g_log == this)
-		g_log = 0;
-}
-
-void rcLog::log(rcLogCategory category, const char* format, ...)
-{
-	if (m_messageCount >= MAX_MESSAGES)
-		return;
-	char* dst = &m_textPool[m_textPoolSize];
-	int n = TEXT_POOL_SIZE - m_textPoolSize;
-	if (n < 2)
-		return;
-	// Store category
-	*dst = (char)category;
-	n--;
-	// Store message
-	va_list ap;
-	va_start(ap, format);
-	int ret = vsnprintf(dst+1, n-1, format, ap);
-	va_end(ap);
-	if (ret > 0)
-		m_textPoolSize += ret+2;
-	m_messages[m_messageCount++] = dst;
-}
-
-void rcSetLog(rcLog* log)
-{
-	g_log = log;
-}
-
-rcLog* rcGetLog()
-{
-	return g_log;
-}
-
-void rcSetBuildTimes(rcBuildTimes* btimes)
-{
-	g_btimes = btimes;
-}
-
-rcBuildTimes* rcGetBuildTimes()
-{
-	return g_btimes;
-}

extern/recastnavigation/Recast/Source/RecastMesh.cpp

@@ -160,6 +160,7 @@ static unsigned short addVertex(unsigned short x, unsigned short y, unsigned sho
 	return (unsigned short)i;
 }
 
+// Last time I checked the if version got compiled using cmov, which was a lot faster than module (with idiv).
 inline int prev(int i, int n) { return i-1 >= 0 ? i-1 : n-1; }
 inline int next(int i, int n) { return i+1 < n ? i+1 : 0; }
 
@@ -288,6 +289,53 @@ static bool diagonal(int i, int j, int n, const int* verts, int* indices)
 	return inCone(i, j, n, verts, indices) && diagonalie(i, j, n, verts, indices);
 }
 
+
+static bool diagonalieLoose(int i, int j, int n, const int* verts, int* indices)
+{
+	const int* d0 = &verts[(indices[i] & 0x0fffffff) * 4];
+	const int* d1 = &verts[(indices[j] & 0x0fffffff) * 4];
+	
+	// For each edge (k,k+1) of P
+	for (int k = 0; k < n; k++)
+	{
+		int k1 = next(k, n);
+		// Skip edges incident to i or j
+		if (!((k == i) || (k1 == i) || (k == j) || (k1 == j)))
+		{
+			const int* p0 = &verts[(indices[k] & 0x0fffffff) * 4];
+			const int* p1 = &verts[(indices[k1] & 0x0fffffff) * 4];
+			
+			if (vequal(d0, p0) || vequal(d1, p0) || vequal(d0, p1) || vequal(d1, p1))
+				continue;
+			
+			if (intersectProp(d0, d1, p0, p1))
+				return false;
+		}
+	}
+	return true;
+}
+
+static bool	inConeLoose(int i, int j, int n, const int* verts, int* indices)
+{
+	const int* pi = &verts[(indices[i] & 0x0fffffff) * 4];
+	const int* pj = &verts[(indices[j] & 0x0fffffff) * 4];
+	const int* pi1 = &verts[(indices[next(i, n)] & 0x0fffffff) * 4];
+	const int* pin1 = &verts[(indices[prev(i, n)] & 0x0fffffff) * 4];
+	
+	// If P[i] is a convex vertex [ i+1 left or on (i-1,i) ].
+	if (leftOn(pin1, pi, pi1))
+		return leftOn(pi, pj, pin1) && leftOn(pj, pi, pi1);
+	// Assume (i-1,i,i+1) not collinear.
+	// else P[i] is reflex.
+	return !(leftOn(pi, pj, pi1) && leftOn(pj, pi, pin1));
+}
+
+static bool diagonalLoose(int i, int j, int n, const int* verts, int* indices)
+{
+	return inConeLoose(i, j, n, verts, indices) && diagonalieLoose(i, j, n, verts, indices);
+}
+
+
 static int triangulate(int n, const int* verts, int* indices, int* tris)
 {
 	int ntris = 0;
@@ -305,7 +353,7 @@ static int triangulate(int n, const int* verts, int* indices, int* tris)
 	while (n > 3)
 	{
 		int minLen = -1;
-		int i_min = -1;
+		int mini = -1;
 		for (int i = 0; i < n; i++)
 		{
 			int i1 = next(i, n);
@@ -321,24 +369,51 @@ static int triangulate(int n, const int* verts, int* indices, int* tris)
 				if (minLen < 0 || len < minLen)
 				{
 					minLen = len;
-					i_min = i;
+					mini = i;
 				}
 			}
 		}
 		
-		if (i_min == -1)
+		if (mini == -1)
 		{
-			// Should not happen.
-/*			printf("mini == -1 ntris=%d n=%d\n", ntris, n);
+			// We might get here because the contour has overlapping segments, like this:
+			//
+			//  A o-o=====o---o B
+			//   /  |C   D|    \
+			//  o   o     o     o
+			//  :   :     :     :
+			// We'll try to recover by loosing up the inCone test a bit so that a diagonal
+			// like A-B or C-D can be found and we can continue.
+			minLen = -1;
+			mini = -1;
 			for (int i = 0; i < n; i++)
 			{
-				printf("%d ", indices[i] & 0x0fffffff);
+				int i1 = next(i, n);
+				int i2 = next(i1, n);
+				if (diagonalLoose(i, i2, n, verts, indices))
+				{
+					const int* p0 = &verts[(indices[i] & 0x0fffffff) * 4];
+					const int* p2 = &verts[(indices[next(i2, n)] & 0x0fffffff) * 4];
+					int dx = p2[0] - p0[0];
+					int dy = p2[2] - p0[2];
+					int len = dx*dx + dy*dy;
+					
+					if (minLen < 0 || len < minLen)
+					{
+						minLen = len;
+						mini = i;
+					}
+				}
+			}
+			if (mini == -1)
+			{
+				// The contour is messed up. This sometimes happens
+				// if the contour simplification is too aggressive.
+				return -ntris;
 			}
-			printf("\n");*/
-			return -ntris;
 		}
 		
-		int i = i_min;
+		int i = mini;
 		int i1 = next(i, n);
 		int i2 = next(i1, n);
 		
@@ -453,8 +528,8 @@ static int getPolyMergeValue(unsigned short* pa, unsigned short* pb,
 	return dx*dx + dy*dy;
 }
 
-static void mergePolys(unsigned short* pa, unsigned short* pb, int ea, int eb,
-					   unsigned short* tmp, const int nvp)
+static void mergePolyVerts(unsigned short* pa, unsigned short* pb, int ea, int eb,
+						   unsigned short* tmp, const int nvp)
 {
 	const int na = countPolyVerts(pa, nvp);
 	const int nb = countPolyVerts(pb, nvp);
@@ -526,7 +601,7 @@ static bool canRemoveVertex(rcContext* ctx, rcPolyMesh& mesh, const unsigned sho
 	// Find edges which share the removed vertex.
 	const int maxEdges = numTouchedVerts*2;
 	int nedges = 0;
-	rcScopedDelete<int> edges = (int*)rcAlloc(sizeof(int)*maxEdges*3, RC_ALLOC_TEMP);
+	rcScopedDelete<int> edges((int*)rcAlloc(sizeof(int)*maxEdges*3, RC_ALLOC_TEMP));
 	if (!edges)
 	{
 		ctx->log(RC_LOG_WARNING, "canRemoveVertex: Out of memory 'edges' (%d).", maxEdges*3);
@@ -550,9 +625,9 @@ static bool canRemoveVertex(rcContext* ctx, rcPolyMesh& mesh, const unsigned sho
 					
 				// Check if the edge exists
 				bool exists = false;
-				for (int k = 0; k < nedges; ++k)
+				for (int m = 0; m < nedges; ++m)
 				{
-					int* e = &edges[k*3];
+					int* e = &edges[m*3];
 					if (e[1] == b)
 					{
 						// Exists, increment vertex share count.
@@ -606,7 +681,7 @@ static bool removeVertex(rcContext* ctx, rcPolyMesh& mesh, const unsigned short
 	}
 	
 	int nedges = 0;
-	rcScopedDelete<int> edges = (int*)rcAlloc(sizeof(int)*numRemovedVerts*nvp*4, RC_ALLOC_TEMP);
+	rcScopedDelete<int> edges((int*)rcAlloc(sizeof(int)*numRemovedVerts*nvp*4, RC_ALLOC_TEMP));
 	if (!edges)
 	{
 		ctx->log(RC_LOG_WARNING, "removeVertex: Out of memory 'edges' (%d).", numRemovedVerts*nvp*4);
@@ -614,15 +689,15 @@ static bool removeVertex(rcContext* ctx, rcPolyMesh& mesh, const unsigned short
 	}
 
 	int nhole = 0;
-	rcScopedDelete<int> hole = (int*)rcAlloc(sizeof(int)*numRemovedVerts*nvp, RC_ALLOC_TEMP);
+	rcScopedDelete<int> hole((int*)rcAlloc(sizeof(int)*numRemovedVerts*nvp, RC_ALLOC_TEMP));
 	if (!hole)
 	{
 		ctx->log(RC_LOG_WARNING, "removeVertex: Out of memory 'hole' (%d).", numRemovedVerts*nvp);
 		return false;
 	}
-	
+
 	int nhreg = 0;
-	rcScopedDelete<int> hreg = (int*)rcAlloc(sizeof(int)*numRemovedVerts*nvp, RC_ALLOC_TEMP);
+	rcScopedDelete<int> hreg((int*)rcAlloc(sizeof(int)*numRemovedVerts*nvp, RC_ALLOC_TEMP));
 	if (!hreg)
 	{
 		ctx->log(RC_LOG_WARNING, "removeVertex: Out of memory 'hreg' (%d).", numRemovedVerts*nvp);
@@ -630,7 +705,7 @@ static bool removeVertex(rcContext* ctx, rcPolyMesh& mesh, const unsigned short
 	}
 
 	int nharea = 0;
-	rcScopedDelete<int> harea = (int*)rcAlloc(sizeof(int)*numRemovedVerts*nvp, RC_ALLOC_TEMP);
+	rcScopedDelete<int> harea((int*)rcAlloc(sizeof(int)*numRemovedVerts*nvp, RC_ALLOC_TEMP));
 	if (!harea)
 	{
 		ctx->log(RC_LOG_WARNING, "removeVertex: Out of memory 'harea' (%d).", numRemovedVerts*nvp);
@@ -661,7 +736,8 @@ static bool removeVertex(rcContext* ctx, rcPolyMesh& mesh, const unsigned short
 			}
 			// Remove the polygon.
 			unsigned short* p2 = &mesh.polys[(mesh.npolys-1)*nvp*2];
-			memcpy(p,p2,sizeof(unsigned short)*nvp);
+			if (p != p2)
+				memcpy(p,p2,sizeof(unsigned short)*nvp);
 			memset(p+nvp,0xff,sizeof(unsigned short)*nvp);
 			mesh.regs[i] = mesh.regs[mesh.npolys-1];
 			mesh.areas[i] = mesh.areas[mesh.npolys-1];
@@ -671,7 +747,7 @@ static bool removeVertex(rcContext* ctx, rcPolyMesh& mesh, const unsigned short
 	}
 	
 	// Remove vertex.
-	for (int i = (int)rem; i < mesh.nverts; ++i)
+	for (int i = (int)rem; i < mesh.nverts - 1; ++i)
 	{
 		mesh.verts[i*3+0] = mesh.verts[(i+1)*3+0];
 		mesh.verts[i*3+1] = mesh.verts[(i+1)*3+1];
@@ -746,22 +822,22 @@ static bool removeVertex(rcContext* ctx, rcPolyMesh& mesh, const unsigned short
 			break;
 	}
 
-	rcScopedDelete<int> tris = (int*)rcAlloc(sizeof(int)*nhole*3, RC_ALLOC_TEMP);
+	rcScopedDelete<int> tris((int*)rcAlloc(sizeof(int)*nhole*3, RC_ALLOC_TEMP));
 	if (!tris)
 	{
 		ctx->log(RC_LOG_WARNING, "removeVertex: Out of memory 'tris' (%d).", nhole*3);
 		return false;
 	}
 
-	rcScopedDelete<int> tverts = (int*)rcAlloc(sizeof(int)*nhole*4, RC_ALLOC_TEMP);
+	rcScopedDelete<int> tverts((int*)rcAlloc(sizeof(int)*nhole*4, RC_ALLOC_TEMP));
 	if (!tverts)
 	{
 		ctx->log(RC_LOG_WARNING, "removeVertex: Out of memory 'tverts' (%d).", nhole*4);
 		return false;
 	}
 
-	rcScopedDelete<int> thole = (int*)rcAlloc(sizeof(int)*nhole, RC_ALLOC_TEMP);
-	if (!tverts)
+	rcScopedDelete<int> thole((int*)rcAlloc(sizeof(int)*nhole, RC_ALLOC_TEMP));
+	if (!thole)
 	{
 		ctx->log(RC_LOG_WARNING, "removeVertex: Out of memory 'thole' (%d).", nhole);
 		return false;
@@ -787,20 +863,20 @@ static bool removeVertex(rcContext* ctx, rcPolyMesh& mesh, const unsigned short
 	}
 	
 	// Merge the hole triangles back to polygons.
-	rcScopedDelete<unsigned short> polys = (unsigned short*)rcAlloc(sizeof(unsigned short)*(ntris+1)*nvp, RC_ALLOC_TEMP);
+	rcScopedDelete<unsigned short> polys((unsigned short*)rcAlloc(sizeof(unsigned short)*(ntris+1)*nvp, RC_ALLOC_TEMP));
 	if (!polys)
 	{
 		ctx->log(RC_LOG_ERROR, "removeVertex: Out of memory 'polys' (%d).", (ntris+1)*nvp);
 		return false;
 	}
-	rcScopedDelete<unsigned short> pregs = (unsigned short*)rcAlloc(sizeof(unsigned short)*ntris, RC_ALLOC_TEMP);
+	rcScopedDelete<unsigned short> pregs((unsigned short*)rcAlloc(sizeof(unsigned short)*ntris, RC_ALLOC_TEMP));
 	if (!pregs)
 	{
 		ctx->log(RC_LOG_ERROR, "removeVertex: Out of memory 'pregs' (%d).", ntris);
 		return false;
 	}
-	rcScopedDelete<unsigned char> pareas = (unsigned char*)rcAlloc(sizeof(unsigned char)*ntris, RC_ALLOC_TEMP);
-	if (!pregs)
+	rcScopedDelete<unsigned char> pareas((unsigned char*)rcAlloc(sizeof(unsigned char)*ntris, RC_ALLOC_TEMP));
+	if (!pareas)
 	{
 		ctx->log(RC_LOG_ERROR, "removeVertex: Out of memory 'pareas' (%d).", ntris);
 		return false;
@@ -819,7 +895,14 @@ static bool removeVertex(rcContext* ctx, rcPolyMesh& mesh, const unsigned short
 			polys[npolys*nvp+0] = (unsigned short)hole[t[0]];
 			polys[npolys*nvp+1] = (unsigned short)hole[t[1]];
 			polys[npolys*nvp+2] = (unsigned short)hole[t[2]];
-			pregs[npolys] = (unsigned short)hreg[t[0]];
+
+			// If this polygon covers multiple region types then
+			// mark it as such
+			if (hreg[t[0]] != hreg[t[1]] || hreg[t[1]] != hreg[t[2]])
+				pregs[npolys] = RC_MULTIPLE_REGS;
+			else
+				pregs[npolys] = (unsigned short)hreg[t[0]];
+
 			pareas[npolys] = (unsigned char)harea[t[0]];
 			npolys++;
 		}
@@ -860,8 +943,13 @@ static bool removeVertex(rcContext* ctx, rcPolyMesh& mesh, const unsigned short
 				// Found best, merge.
 				unsigned short* pa = &polys[bestPa*nvp];
 				unsigned short* pb = &polys[bestPb*nvp];
-				mergePolys(pa, pb, bestEa, bestEb, tmpPoly, nvp);
-				memcpy(pb, &polys[(npolys-1)*nvp], sizeof(unsigned short)*nvp);
+				mergePolyVerts(pa, pb, bestEa, bestEb, tmpPoly, nvp);
+				if (pregs[bestPa] != pregs[bestPb])
+					pregs[bestPa] = RC_MULTIPLE_REGS;
+
+				unsigned short* last = &polys[(npolys-1)*nvp];
+				if (pb != last)
+					memcpy(pb, last, sizeof(unsigned short)*nvp);
 				pregs[bestPb] = pregs[npolys-1];
 				pareas[bestPb] = pareas[npolys-1];
 				npolys--;
@@ -905,13 +993,14 @@ bool rcBuildPolyMesh(rcContext* ctx, rcContourSet& cset, const int nvp, rcPolyMe
 {
 	rcAssert(ctx);
 	
-	ctx->startTimer(RC_TIMER_BUILD_POLYMESH);
+	rcScopedTimer timer(ctx, RC_TIMER_BUILD_POLYMESH);
 
 	rcVcopy(mesh.bmin, cset.bmin);
 	rcVcopy(mesh.bmax, cset.bmax);
 	mesh.cs = cset.cs;
 	mesh.ch = cset.ch;
 	mesh.borderSize = cset.borderSize;
+	mesh.maxEdgeError = cset.maxError;
 	
 	int maxVertices = 0;
 	int maxTris = 0;
@@ -931,10 +1020,10 @@ bool rcBuildPolyMesh(rcContext* ctx, rcContourSet& cset, const int nvp, rcPolyMe
 		return false;
 	}
 		
-	rcScopedDelete<unsigned char> vflags = (unsigned char*)rcAlloc(sizeof(unsigned char)*maxVertices, RC_ALLOC_TEMP);
+	rcScopedDelete<unsigned char> vflags((unsigned char*)rcAlloc(sizeof(unsigned char)*maxVertices, RC_ALLOC_TEMP));
 	if (!vflags)
 	{
-		ctx->log(RC_LOG_ERROR, "rcBuildPolyMesh: Out of memory 'mesh.verts' (%d).", maxVertices);
+		ctx->log(RC_LOG_ERROR, "rcBuildPolyMesh: Out of memory 'vflags' (%d).", maxVertices);
 		return false;
 	}
 	memset(vflags, 0, maxVertices);
@@ -974,7 +1063,7 @@ bool rcBuildPolyMesh(rcContext* ctx, rcContourSet& cset, const int nvp, rcPolyMe
 	memset(mesh.regs, 0, sizeof(unsigned short)*maxTris);
 	memset(mesh.areas, 0, sizeof(unsigned char)*maxTris);
 	
-	rcScopedDelete<int> nextVert = (int*)rcAlloc(sizeof(int)*maxVertices, RC_ALLOC_TEMP);
+	rcScopedDelete<int> nextVert((int*)rcAlloc(sizeof(int)*maxVertices, RC_ALLOC_TEMP));
 	if (!nextVert)
 	{
 		ctx->log(RC_LOG_ERROR, "rcBuildPolyMesh: Out of memory 'nextVert' (%d).", maxVertices);
@@ -982,7 +1071,7 @@ bool rcBuildPolyMesh(rcContext* ctx, rcContourSet& cset, const int nvp, rcPolyMe
 	}
 	memset(nextVert, 0, sizeof(int)*maxVertices);
 	
-	rcScopedDelete<int> firstVert = (int*)rcAlloc(sizeof(int)*VERTEX_BUCKET_COUNT, RC_ALLOC_TEMP);
+	rcScopedDelete<int> firstVert((int*)rcAlloc(sizeof(int)*VERTEX_BUCKET_COUNT, RC_ALLOC_TEMP));
 	if (!firstVert)
 	{
 		ctx->log(RC_LOG_ERROR, "rcBuildPolyMesh: Out of memory 'firstVert' (%d).", VERTEX_BUCKET_COUNT);
@@ -991,19 +1080,19 @@ bool rcBuildPolyMesh(rcContext* ctx, rcContourSet& cset, const int nvp, rcPolyMe
 	for (int i = 0; i < VERTEX_BUCKET_COUNT; ++i)
 		firstVert[i] = -1;
 	
-	rcScopedDelete<int> indices = (int*)rcAlloc(sizeof(int)*maxVertsPerCont, RC_ALLOC_TEMP);
+	rcScopedDelete<int> indices((int*)rcAlloc(sizeof(int)*maxVertsPerCont, RC_ALLOC_TEMP));
 	if (!indices)
 	{
 		ctx->log(RC_LOG_ERROR, "rcBuildPolyMesh: Out of memory 'indices' (%d).", maxVertsPerCont);
 		return false;
 	}
-	rcScopedDelete<int> tris = (int*)rcAlloc(sizeof(int)*maxVertsPerCont*3, RC_ALLOC_TEMP);
+	rcScopedDelete<int> tris((int*)rcAlloc(sizeof(int)*maxVertsPerCont*3, RC_ALLOC_TEMP));
 	if (!tris)
 	{
 		ctx->log(RC_LOG_ERROR, "rcBuildPolyMesh: Out of memory 'tris' (%d).", maxVertsPerCont*3);
 		return false;
 	}
-	rcScopedDelete<unsigned short> polys = (unsigned short*)rcAlloc(sizeof(unsigned short)*(maxVertsPerCont+1)*nvp, RC_ALLOC_TEMP);
+	rcScopedDelete<unsigned short> polys((unsigned short*)rcAlloc(sizeof(unsigned short)*(maxVertsPerCont+1)*nvp, RC_ALLOC_TEMP));
 	if (!polys)
 	{
 		ctx->log(RC_LOG_ERROR, "rcBuildPolyMesh: Out of memory 'polys' (%d).", maxVertsPerCont*nvp);
@@ -1053,7 +1142,7 @@ bool rcBuildPolyMesh(rcContext* ctx, rcContourSet& cset, const int nvp, rcPolyMe
 				vflags[indices[j]] = 1;
 			}
 		}
-		
+
 		// Build initial polygons.
 		int npolys = 0;
 		memset(polys, 0xff, maxVertsPerCont*nvp*sizeof(unsigned short));
@@ -1104,8 +1193,10 @@ bool rcBuildPolyMesh(rcContext* ctx, rcContourSet& cset, const int nvp, rcPolyMe
 					// Found best, merge.
 					unsigned short* pa = &polys[bestPa*nvp];
 					unsigned short* pb = &polys[bestPb*nvp];
-					mergePolys(pa, pb, bestEa, bestEb, tmpPoly, nvp);
-					memcpy(pb, &polys[(npolys-1)*nvp], sizeof(unsigned short)*nvp);
+					mergePolyVerts(pa, pb, bestEa, bestEb, tmpPoly, nvp);
+					unsigned short* lastPoly = &polys[(npolys-1)*nvp];
+					if (pb != lastPoly)
+						memcpy(pb, lastPoly, sizeof(unsigned short)*nvp);
 					npolys--;
 				}
 				else
@@ -1150,6 +1241,7 @@ bool rcBuildPolyMesh(rcContext* ctx, rcContourSet& cset, const int nvp, rcPolyMe
 			}
 			// Remove vertex
 			// Note: mesh.nverts is already decremented inside removeVertex()!
+			// Fixup vertex flags
 			for (int j = i; j < mesh.nverts; ++j)
 				vflags[j] = vflags[j+1];
 			--i;
@@ -1212,8 +1304,6 @@ bool rcBuildPolyMesh(rcContext* ctx, rcContourSet& cset, const int nvp, rcPolyMe
 		ctx->log(RC_LOG_ERROR, "rcBuildPolyMesh: The resulting mesh has too many polygons %d (max %d). Data can be corrupted.", mesh.npolys, 0xffff);
 	}
 	
-	ctx->stopTimer(RC_TIMER_BUILD_POLYMESH);
-	
 	return true;
 }
 
@@ -1225,7 +1315,7 @@ bool rcMergePolyMeshes(rcContext* ctx, rcPolyMesh** meshes, const int nmeshes, r
 	if (!nmeshes || !meshes)
 		return true;
 
-	ctx->startTimer(RC_TIMER_MERGE_POLYMESH);
+	rcScopedTimer timer(ctx, RC_TIMER_MERGE_POLYMESH);
 
 	mesh.nvp = meshes[0]->nvp;
 	mesh.cs = meshes[0]->cs;
@@ -1286,7 +1376,7 @@ bool rcMergePolyMeshes(rcContext* ctx, rcPolyMesh** meshes, const int nmeshes, r
 	}
 	memset(mesh.flags, 0, sizeof(unsigned short)*maxPolys);
 	
-	rcScopedDelete<int> nextVert = (int*)rcAlloc(sizeof(int)*maxVerts, RC_ALLOC_TEMP);
+	rcScopedDelete<int> nextVert((int*)rcAlloc(sizeof(int)*maxVerts, RC_ALLOC_TEMP));
 	if (!nextVert)
 	{
 		ctx->log(RC_LOG_ERROR, "rcMergePolyMeshes: Out of memory 'nextVert' (%d).", maxVerts);
@@ -1294,7 +1384,7 @@ bool rcMergePolyMeshes(rcContext* ctx, rcPolyMesh** meshes, const int nmeshes, r
 	}
 	memset(nextVert, 0, sizeof(int)*maxVerts);
 	
-	rcScopedDelete<int> firstVert = (int*)rcAlloc(sizeof(int)*VERTEX_BUCKET_COUNT, RC_ALLOC_TEMP);
+	rcScopedDelete<int> firstVert((int*)rcAlloc(sizeof(int)*VERTEX_BUCKET_COUNT, RC_ALLOC_TEMP));
 	if (!firstVert)
 	{
 		ctx->log(RC_LOG_ERROR, "rcMergePolyMeshes: Out of memory 'firstVert' (%d).", VERTEX_BUCKET_COUNT);
@@ -1303,13 +1393,13 @@ bool rcMergePolyMeshes(rcContext* ctx, rcPolyMesh** meshes, const int nmeshes, r
 	for (int i = 0; i < VERTEX_BUCKET_COUNT; ++i)
 		firstVert[i] = -1;
 
-	rcScopedDelete<unsigned short> vremap = (unsigned short*)rcAlloc(sizeof(unsigned short)*maxVertsPerMesh, RC_ALLOC_PERM);
+	rcScopedDelete<unsigned short> vremap((unsigned short*)rcAlloc(sizeof(unsigned short)*maxVertsPerMesh, RC_ALLOC_PERM));
 	if (!vremap)
 	{
 		ctx->log(RC_LOG_ERROR, "rcMergePolyMeshes: Out of memory 'vremap' (%d).", maxVertsPerMesh);
 		return false;
 	}
-	memset(nextVert, 0, sizeof(int)*maxVerts);
+	memset(vremap, 0, sizeof(unsigned short)*maxVertsPerMesh);
 	
 	for (int i = 0; i < nmeshes; ++i)
 	{
@@ -1318,6 +1408,12 @@ bool rcMergePolyMeshes(rcContext* ctx, rcPolyMesh** meshes, const int nmeshes, r
 		const unsigned short ox = (unsigned short)floorf((pmesh->bmin[0]-mesh.bmin[0])/mesh.cs+0.5f);
 		const unsigned short oz = (unsigned short)floorf((pmesh->bmin[2]-mesh.bmin[2])/mesh.cs+0.5f);
 		
+		bool isMinX = (ox == 0);
+		bool isMinZ = (oz == 0);
+		bool isMaxX = ((unsigned short)floorf((mesh.bmax[0] - pmesh->bmax[0]) / mesh.cs + 0.5f)) == 0;
+		bool isMaxZ = ((unsigned short)floorf((mesh.bmax[2] - pmesh->bmax[2]) / mesh.cs + 0.5f)) == 0;
+		bool isOnBorder = (isMinX || isMinZ || isMaxX || isMaxZ);
+
 		for (int j = 0; j < pmesh->nverts; ++j)
 		{
 			unsigned short* v = &pmesh->verts[j*3];
@@ -1338,6 +1434,36 @@ bool rcMergePolyMeshes(rcContext* ctx, rcPolyMesh** meshes, const int nmeshes, r
 				if (src[k] == RC_MESH_NULL_IDX) break;
 				tgt[k] = vremap[src[k]];
 			}
+
+			if (isOnBorder)
+			{
+				for (int k = mesh.nvp; k < mesh.nvp * 2; ++k)
+				{
+					if (src[k] & 0x8000 && src[k] != 0xffff)
+					{
+						unsigned short dir = src[k] & 0xf;
+						switch (dir)
+						{
+							case 0: // Portal x-
+								if (isMinX)
+									tgt[k] = src[k];
+								break;
+							case 1: // Portal z+
+								if (isMaxZ)
+									tgt[k] = src[k];
+								break;
+							case 2: // Portal x+
+								if (isMaxX)
+									tgt[k] = src[k];
+								break;
+							case 3: // Portal z-
+								if (isMinZ)
+									tgt[k] = src[k];
+								break;
+						}
+					}
+				}
+			}
 		}
 	}
 
@@ -1357,7 +1483,70 @@ bool rcMergePolyMeshes(rcContext* ctx, rcPolyMesh** meshes, const int nmeshes, r
 		ctx->log(RC_LOG_ERROR, "rcMergePolyMeshes: The resulting mesh has too many polygons %d (max %d). Data can be corrupted.", mesh.npolys, 0xffff);
 	}
 	
-	ctx->stopTimer(RC_TIMER_MERGE_POLYMESH);
+	return true;
+}
+
+bool rcCopyPolyMesh(rcContext* ctx, const rcPolyMesh& src, rcPolyMesh& dst)
+{
+	rcAssert(ctx);
+	
+	// Destination must be empty.
+	rcAssert(dst.verts == 0);
+	rcAssert(dst.polys == 0);
+	rcAssert(dst.regs == 0);
+	rcAssert(dst.areas == 0);
+	rcAssert(dst.flags == 0);
+	
+	dst.nverts = src.nverts;
+	dst.npolys = src.npolys;
+	dst.maxpolys = src.npolys;
+	dst.nvp = src.nvp;
+	rcVcopy(dst.bmin, src.bmin);
+	rcVcopy(dst.bmax, src.bmax);
+	dst.cs = src.cs;
+	dst.ch = src.ch;
+	dst.borderSize = src.borderSize;
+	dst.maxEdgeError = src.maxEdgeError;
+	
+	dst.verts = (unsigned short*)rcAlloc(sizeof(unsigned short)*src.nverts*3, RC_ALLOC_PERM);
+	if (!dst.verts)
+	{
+		ctx->log(RC_LOG_ERROR, "rcCopyPolyMesh: Out of memory 'dst.verts' (%d).", src.nverts*3);
+		return false;
+	}
+	memcpy(dst.verts, src.verts, sizeof(unsigned short)*src.nverts*3);
+	
+	dst.polys = (unsigned short*)rcAlloc(sizeof(unsigned short)*src.npolys*2*src.nvp, RC_ALLOC_PERM);
+	if (!dst.polys)
+	{
+		ctx->log(RC_LOG_ERROR, "rcCopyPolyMesh: Out of memory 'dst.polys' (%d).", src.npolys*2*src.nvp);
+		return false;
+	}
+	memcpy(dst.polys, src.polys, sizeof(unsigned short)*src.npolys*2*src.nvp);
+	
+	dst.regs = (unsigned short*)rcAlloc(sizeof(unsigned short)*src.npolys, RC_ALLOC_PERM);
+	if (!dst.regs)
+	{
+		ctx->log(RC_LOG_ERROR, "rcCopyPolyMesh: Out of memory 'dst.regs' (%d).", src.npolys);
+		return false;
+	}
+	memcpy(dst.regs, src.regs, sizeof(unsigned short)*src.npolys);
+	
+	dst.areas = (unsigned char*)rcAlloc(sizeof(unsigned char)*src.npolys, RC_ALLOC_PERM);
+	if (!dst.areas)
+	{
+		ctx->log(RC_LOG_ERROR, "rcCopyPolyMesh: Out of memory 'dst.areas' (%d).", src.npolys);
+		return false;
+	}
+	memcpy(dst.areas, src.areas, sizeof(unsigned char)*src.npolys);
+	
+	dst.flags = (unsigned short*)rcAlloc(sizeof(unsigned short)*src.npolys, RC_ALLOC_PERM);
+	if (!dst.flags)
+	{
+		ctx->log(RC_LOG_ERROR, "rcCopyPolyMesh: Out of memory 'dst.flags' (%d).", src.npolys);
+		return false;
+	}
+	memcpy(dst.flags, src.flags, sizeof(unsigned short)*src.npolys);
 	
 	return true;
 }

extern/recastnavigation/Recast/Source/RecastRasterization.cpp

@@ -50,7 +50,7 @@ static rcSpan* allocSpan(rcHeightfield& hf)
 		// Allocate memory for the new pool.
 		rcSpanPool* pool = (rcSpanPool*)rcAlloc(sizeof(rcSpanPool), RC_ALLOC_PERM);
 		if (!pool) return 0;
-		pool->next = 0;
+
 		// Add the pool into the list of pools.
 		pool->next = hf.pools;
 		hf.pools = pool;
@@ -82,7 +82,7 @@ static void freeSpan(rcHeightfield& hf, rcSpan* ptr)
 	hf.freelist = ptr;
 }
 
-static void addSpan(rcHeightfield& hf, const int x, const int y,
+static bool addSpan(rcHeightfield& hf, const int x, const int y,
 					const unsigned short smin, const unsigned short smax,
 					const unsigned char area, const int flagMergeThr)
 {
@@ -90,16 +90,18 @@ static void addSpan(rcHeightfield& hf, const int x, const int y,
 	int idx = x + y*hf.width;
 	
 	rcSpan* s = allocSpan(hf);
+	if (!s)
+		return false;
 	s->smin = smin;
 	s->smax = smax;
 	s->area = area;
 	s->next = 0;
 	
-	// Empty cell, add he first span.
+	// Empty cell, add the first span.
 	if (!hf.spans[idx])
 	{
 		hf.spans[idx] = s;
-		return;
+		return true;
 	}
 	rcSpan* prev = 0;
 	rcSpan* cur = hf.spans[idx];
@@ -152,6 +154,8 @@ static void addSpan(rcHeightfield& hf, const int x, const int y,
 		s->next = hf.spans[idx];
 		hf.spans[idx] = s;
 	}
+
+	return true;
 }
 
 /// @par
@@ -161,45 +165,80 @@ static void addSpan(rcHeightfield& hf, const int x, const int y,
 /// from the existing span, the span flags are merged.
 ///
 /// @see rcHeightfield, rcSpan.
-void rcAddSpan(rcContext* /*ctx*/, rcHeightfield& hf, const int x, const int y,
+bool rcAddSpan(rcContext* ctx, rcHeightfield& hf, const int x, const int y,
 			   const unsigned short smin, const unsigned short smax,
 			   const unsigned char area, const int flagMergeThr)
 {
-//	rcAssert(ctx);
-	addSpan(hf, x,y, smin, smax, area, flagMergeThr);
+	rcAssert(ctx);
+
+	if (!addSpan(hf, x, y, smin, smax, area, flagMergeThr))
+	{
+		ctx->log(RC_LOG_ERROR, "rcAddSpan: Out of memory.");
+		return false;
+	}
+
+	return true;
 }
 
-static int clipPoly(const float* in, int n, float* out, float pnx, float pnz, float pd)
+// divides a convex polygons into two convex polygons on both sides of a line
+static void dividePoly(const float* in, int nin,
+					  float* out1, int* nout1,
+					  float* out2, int* nout2,
+					  float x, int axis)
 {
 	float d[12];
-	for (int i = 0; i < n; ++i)
-		d[i] = pnx*in[i*3+0] + pnz*in[i*3+2] + pd;
-	
-	int m = 0;
-	for (int i = 0, j = n-1; i < n; j=i, ++i)
+	for (int i = 0; i < nin; ++i)
+		d[i] = x - in[i*3+axis];
+
+	int m = 0, n = 0;
+	for (int i = 0, j = nin-1; i < nin; j=i, ++i)
 	{
 		bool ina = d[j] >= 0;
 		bool inb = d[i] >= 0;
 		if (ina != inb)
 		{
 			float s = d[j] / (d[j] - d[i]);
-			out[m*3+0] = in[j*3+0] + (in[i*3+0] - in[j*3+0])*s;
-			out[m*3+1] = in[j*3+1] + (in[i*3+1] - in[j*3+1])*s;
-			out[m*3+2] = in[j*3+2] + (in[i*3+2] - in[j*3+2])*s;
+			out1[m*3+0] = in[j*3+0] + (in[i*3+0] - in[j*3+0])*s;
+			out1[m*3+1] = in[j*3+1] + (in[i*3+1] - in[j*3+1])*s;
+			out1[m*3+2] = in[j*3+2] + (in[i*3+2] - in[j*3+2])*s;
+			rcVcopy(out2 + n*3, out1 + m*3);
 			m++;
+			n++;
+			// add the i'th point to the right polygon. Do NOT add points that are on the dividing line
+			// since these were already added above
+			if (d[i] > 0)
+			{
+				rcVcopy(out1 + m*3, in + i*3);
+				m++;
+			}
+			else if (d[i] < 0)
+			{
+				rcVcopy(out2 + n*3, in + i*3);
+				n++;
+			}
 		}
-		if (inb)
+		else // same side
 		{
-			out[m*3+0] = in[i*3+0];
-			out[m*3+1] = in[i*3+1];
-			out[m*3+2] = in[i*3+2];
-			m++;
+			// add the i'th point to the right polygon. Addition is done even for points on the dividing line
+			if (d[i] >= 0)
+			{
+				rcVcopy(out1 + m*3, in + i*3);
+				m++;
+				if (d[i] != 0)
+					continue;
+			}
+			rcVcopy(out2 + n*3, in + i*3);
+			n++;
 		}
 	}
-	return m;
+
+	*nout1 = m;
+	*nout2 = n;
 }
 
-static void rasterizeTri(const float* v0, const float* v1, const float* v2,
+
+
+static bool rasterizeTri(const float* v0, const float* v1, const float* v2,
 						 const unsigned char area, rcHeightfield& hf,
 						 const float* bmin, const float* bmax,
 						 const float cs, const float ics, const float ich,
@@ -220,50 +259,59 @@ static void rasterizeTri(const float* v0, const float* v1, const float* v2,
 	
 	// If the triangle does not touch the bbox of the heightfield, skip the triagle.
 	if (!overlapBounds(bmin, bmax, tmin, tmax))
-		return;
+		return true;
 	
-	// Calculate the footpring of the triangle on the grid.
-	int x0 = (int)((tmin[0] - bmin[0])*ics);
+	// Calculate the footprint of the triangle on the grid's y-axis
 	int y0 = (int)((tmin[2] - bmin[2])*ics);
-	int x1 = (int)((tmax[0] - bmin[0])*ics);
 	int y1 = (int)((tmax[2] - bmin[2])*ics);
-	x0 = rcClamp(x0, 0, w-1);
 	y0 = rcClamp(y0, 0, h-1);
-	x1 = rcClamp(x1, 0, w-1);
 	y1 = rcClamp(y1, 0, h-1);
 	
 	// Clip the triangle into all grid cells it touches.
-	float in[7*3], out[7*3], inrow[7*3];
+	float buf[7*3*4];
+	float *in = buf, *inrow = buf+7*3, *p1 = inrow+7*3, *p2 = p1+7*3;
+
+	rcVcopy(&in[0], v0);
+	rcVcopy(&in[1*3], v1);
+	rcVcopy(&in[2*3], v2);
+	int nvrow, nvIn = 3;
 	
 	for (int y = y0; y <= y1; ++y)
 	{
-		// Clip polygon to row.
-		rcVcopy(&in[0], v0);
-		rcVcopy(&in[1*3], v1);
-		rcVcopy(&in[2*3], v2);
-		int nvrow = 3;
+		// Clip polygon to row. Store the remaining polygon as well
 		const float cz = bmin[2] + y*cs;
-		nvrow = clipPoly(in, nvrow, out, 0, 1, -cz);
-		if (nvrow < 3) continue;
-		nvrow = clipPoly(out, nvrow, inrow, 0, -1, cz+cs);
+		dividePoly(in, nvIn, inrow, &nvrow, p1, &nvIn, cz+cs, 2);
+		rcSwap(in, p1);
 		if (nvrow < 3) continue;
 		
+		// find the horizontal bounds in the row
+		float minX = inrow[0], maxX = inrow[0];
+		for (int i=1; i<nvrow; ++i)
+		{
+			if (minX > inrow[i*3])	minX = inrow[i*3];
+			if (maxX < inrow[i*3])	maxX = inrow[i*3];
+		}
+		int x0 = (int)((minX - bmin[0])*ics);
+		int x1 = (int)((maxX - bmin[0])*ics);
+		x0 = rcClamp(x0, 0, w-1);
+		x1 = rcClamp(x1, 0, w-1);
+
+		int nv, nv2 = nvrow;
+
 		for (int x = x0; x <= x1; ++x)
 		{
-			// Clip polygon to column.
-			int nv = nvrow;
+			// Clip polygon to column. store the remaining polygon as well
 			const float cx = bmin[0] + x*cs;
-			nv = clipPoly(inrow, nv, out, 1, 0, -cx);
-			if (nv < 3) continue;
-			nv = clipPoly(out, nv, in, -1, 0, cx+cs);
+			dividePoly(inrow, nv2, p1, &nv, p2, &nv2, cx+cs, 0);
+			rcSwap(inrow, p2);
 			if (nv < 3) continue;
 			
 			// Calculate min and max of the span.
-			float smin = in[1], smax = in[1];
+			float smin = p1[1], smax = p1[1];
 			for (int i = 1; i < nv; ++i)
 			{
-				smin = rcMin(smin, in[i*3+1]);
-				smax = rcMax(smax, in[i*3+1]);
+				smin = rcMin(smin, p1[i*3+1]);
+				smax = rcMax(smax, p1[i*3+1]);
 			}
 			smin -= bmin[1];
 			smax -= bmin[1];
@@ -278,9 +326,12 @@ static void rasterizeTri(const float* v0, const float* v1, const float* v2,
 			unsigned short ismin = (unsigned short)rcClamp((int)floorf(smin * ich), 0, RC_SPAN_MAX_HEIGHT);
 			unsigned short ismax = (unsigned short)rcClamp((int)ceilf(smax * ich), (int)ismin+1, RC_SPAN_MAX_HEIGHT);
 			
-			addSpan(hf, x, y, ismin, ismax, area, flagMergeThr);
+			if (!addSpan(hf, x, y, ismin, ismax, area, flagMergeThr))
+				return false;
 		}
 	}
+
+	return true;
 }
 
 /// @par
@@ -288,19 +339,23 @@ static void rasterizeTri(const float* v0, const float* v1, const float* v2,
 /// No spans will be added if the triangle does not overlap the heightfield grid.
 ///
 /// @see rcHeightfield
-void rcRasterizeTriangle(rcContext* ctx, const float* v0, const float* v1, const float* v2,
+bool rcRasterizeTriangle(rcContext* ctx, const float* v0, const float* v1, const float* v2,
 						 const unsigned char area, rcHeightfield& solid,
 						 const int flagMergeThr)
 {
 	rcAssert(ctx);
 
-	ctx->startTimer(RC_TIMER_RASTERIZE_TRIANGLES);
+	rcScopedTimer timer(ctx, RC_TIMER_RASTERIZE_TRIANGLES);
 
 	const float ics = 1.0f/solid.cs;
 	const float ich = 1.0f/solid.ch;
-	rasterizeTri(v0, v1, v2, area, solid, solid.bmin, solid.bmax, solid.cs, ics, ich, flagMergeThr);
+	if (!rasterizeTri(v0, v1, v2, area, solid, solid.bmin, solid.bmax, solid.cs, ics, ich, flagMergeThr))
+	{
+		ctx->log(RC_LOG_ERROR, "rcRasterizeTriangle: Out of memory.");
+		return false;
+	}
 
-	ctx->stopTimer(RC_TIMER_RASTERIZE_TRIANGLES);
+	return true;
 }
 
 /// @par
@@ -308,13 +363,13 @@ void rcRasterizeTriangle(rcContext* ctx, const float* v0, const float* v1, const
 /// Spans will only be added for triangles that overlap the heightfield grid.
 ///
 /// @see rcHeightfield
-void rcRasterizeTriangles(rcContext* ctx, const float* verts, const int /*nv*/,
+bool rcRasterizeTriangles(rcContext* ctx, const float* verts, const int /*nv*/,
 						  const int* tris, const unsigned char* areas, const int nt,
 						  rcHeightfield& solid, const int flagMergeThr)
 {
 	rcAssert(ctx);
 
-	ctx->startTimer(RC_TIMER_RASTERIZE_TRIANGLES);
+	rcScopedTimer timer(ctx, RC_TIMER_RASTERIZE_TRIANGLES);
 	
 	const float ics = 1.0f/solid.cs;
 	const float ich = 1.0f/solid.ch;
@@ -325,10 +380,14 @@ void rcRasterizeTriangles(rcContext* ctx, const float* verts, const int /*nv*/,
 		const float* v1 = &verts[tris[i*3+1]*3];
 		const float* v2 = &verts[tris[i*3+2]*3];
 		// Rasterize.
-		rasterizeTri(v0, v1, v2, areas[i], solid, solid.bmin, solid.bmax, solid.cs, ics, ich, flagMergeThr);
+		if (!rasterizeTri(v0, v1, v2, areas[i], solid, solid.bmin, solid.bmax, solid.cs, ics, ich, flagMergeThr))
+		{
+			ctx->log(RC_LOG_ERROR, "rcRasterizeTriangles: Out of memory.");
+			return false;
+		}
 	}
-	
-	ctx->stopTimer(RC_TIMER_RASTERIZE_TRIANGLES);
+
+	return true;
 }
 
 /// @par
@@ -336,13 +395,13 @@ void rcRasterizeTriangles(rcContext* ctx, const float* verts, const int /*nv*/,
 /// Spans will only be added for triangles that overlap the heightfield grid.
 ///
 /// @see rcHeightfield
-void rcRasterizeTriangles(rcContext* ctx, const float* verts, const int /*nv*/,
+bool rcRasterizeTriangles(rcContext* ctx, const float* verts, const int /*nv*/,
 						  const unsigned short* tris, const unsigned char* areas, const int nt,
 						  rcHeightfield& solid, const int flagMergeThr)
 {
 	rcAssert(ctx);
 
-	ctx->startTimer(RC_TIMER_RASTERIZE_TRIANGLES);
+	rcScopedTimer timer(ctx, RC_TIMER_RASTERIZE_TRIANGLES);
 	
 	const float ics = 1.0f/solid.cs;
 	const float ich = 1.0f/solid.ch;
@@ -353,10 +412,14 @@ void rcRasterizeTriangles(rcContext* ctx, const float* verts, const int /*nv*/,
 		const float* v1 = &verts[tris[i*3+1]*3];
 		const float* v2 = &verts[tris[i*3+2]*3];
 		// Rasterize.
-		rasterizeTri(v0, v1, v2, areas[i], solid, solid.bmin, solid.bmax, solid.cs, ics, ich, flagMergeThr);
+		if (!rasterizeTri(v0, v1, v2, areas[i], solid, solid.bmin, solid.bmax, solid.cs, ics, ich, flagMergeThr))
+		{
+			ctx->log(RC_LOG_ERROR, "rcRasterizeTriangles: Out of memory.");
+			return false;
+		}
 	}
-	
-	ctx->stopTimer(RC_TIMER_RASTERIZE_TRIANGLES);
+
+	return true;
 }
 
 /// @par
@@ -364,12 +427,12 @@ void rcRasterizeTriangles(rcContext* ctx, const float* verts, const int /*nv*/,
 /// Spans will only be added for triangles that overlap the heightfield grid.
 ///
 /// @see rcHeightfield
-void rcRasterizeTriangles(rcContext* ctx, const float* verts, const unsigned char* areas, const int nt,
+bool rcRasterizeTriangles(rcContext* ctx, const float* verts, const unsigned char* areas, const int nt,
 						  rcHeightfield& solid, const int flagMergeThr)
 {
 	rcAssert(ctx);
 	
-	ctx->startTimer(RC_TIMER_RASTERIZE_TRIANGLES);
+	rcScopedTimer timer(ctx, RC_TIMER_RASTERIZE_TRIANGLES);
 	
 	const float ics = 1.0f/solid.cs;
 	const float ich = 1.0f/solid.ch;
@@ -380,8 +443,12 @@ void rcRasterizeTriangles(rcContext* ctx, const float* verts, const unsigned cha
 		const float* v1 = &verts[(i*3+1)*3];
 		const float* v2 = &verts[(i*3+2)*3];
 		// Rasterize.
-		rasterizeTri(v0, v1, v2, areas[i], solid, solid.bmin, solid.bmax, solid.cs, ics, ich, flagMergeThr);
+		if (!rasterizeTri(v0, v1, v2, areas[i], solid, solid.bmin, solid.bmax, solid.cs, ics, ich, flagMergeThr))
+		{
+			ctx->log(RC_LOG_ERROR, "rcRasterizeTriangles: Out of memory.");
+			return false;
+		}
 	}
-	
-	ctx->stopTimer(RC_TIMER_RASTERIZE_TRIANGLES);
+
+	return true;
 }

extern/recastnavigation/Recast/Source/RecastRegion.cpp

@@ -286,7 +286,10 @@ static bool floodRegion(int x, int y, int i,
 				if (nr & RC_BORDER_REG) // Do not take borders into account.
 					continue;
 				if (nr != 0 && nr != r)
+				{
 					ar = nr;
+					break;
+				}
 				
 				const rcCompactSpan& as = chf.spans[ai];
 				
@@ -298,9 +301,12 @@ static bool floodRegion(int x, int y, int i,
 					const int ai2 = (int)chf.cells[ax2+ay2*w].index + rcGetCon(as, dir2);
 					if (chf.areas[ai2] != area)
 						continue;
-					unsigned short nr = srcReg[ai2];
-					if (nr != 0 && nr != r)
-						ar = nr;
+					unsigned short nr2 = srcReg[ai2];
+					if (nr2 != 0 && nr2 != r)
+					{
+						ar = nr2;
+						break;
+					}
 				}				
 			}
 		}
@@ -309,6 +315,7 @@ static bool floodRegion(int x, int y, int i,
 			srcReg[ci] = 0;
 			continue;
 		}
+		
 		count++;
 		
 		// Expand neighbours.
@@ -340,30 +347,44 @@ static unsigned short* expandRegions(int maxIter, unsigned short level,
 									 rcCompactHeightfield& chf,
 									 unsigned short* srcReg, unsigned short* srcDist,
 									 unsigned short* dstReg, unsigned short* dstDist, 
-									 rcIntArray& stack)
+									 rcIntArray& stack,
+									 bool fillStack)
 {
 	const int w = chf.width;
 	const int h = chf.height;
 
-	// Find cells revealed by the raised level.
-	stack.resize(0);
-	for (int y = 0; y < h; ++y)
+	if (fillStack)
 	{
-		for (int x = 0; x < w; ++x)
+		// Find cells revealed by the raised level.
+		stack.resize(0);
+		for (int y = 0; y < h; ++y)
 		{
-			const rcCompactCell& c = chf.cells[x+y*w];
-			for (int i = (int)c.index, ni = (int)(c.index+c.count); i < ni; ++i)
+			for (int x = 0; x < w; ++x)
 			{
-				if (chf.dist[i] >= level && srcReg[i] == 0 && chf.areas[i] != RC_NULL_AREA)
+				const rcCompactCell& c = chf.cells[x+y*w];
+				for (int i = (int)c.index, ni = (int)(c.index+c.count); i < ni; ++i)
 				{
-					stack.push(x);
-					stack.push(y);
-					stack.push(i);
+					if (chf.dist[i] >= level && srcReg[i] == 0 && chf.areas[i] != RC_NULL_AREA)
+					{
+						stack.push(x);
+						stack.push(y);
+						stack.push(i);
+					}
 				}
 			}
 		}
 	}
-	
+	else // use cells in the input stack
+	{
+		// mark all cells which already have a region
+		for (int j=0; j<stack.size(); j+=3)
+		{
+			int i = stack[j+2];
+			if (srcReg[i] != 0)
+				stack[j+2] = -1;
+		}
+	}
+
 	int iter = 0;
 	while (stack.size() > 0)
 	{
@@ -434,6 +455,61 @@ static unsigned short* expandRegions(int maxIter, unsigned short level,
 }
 
 
+
+static void sortCellsByLevel(unsigned short startLevel,
+							  rcCompactHeightfield& chf,
+							  unsigned short* srcReg,
+							  unsigned int nbStacks, rcIntArray* stacks,
+							  unsigned short loglevelsPerStack) // the levels per stack (2 in our case) as a bit shift
+{
+	const int w = chf.width;
+	const int h = chf.height;
+	startLevel = startLevel >> loglevelsPerStack;
+
+	for (unsigned int j=0; j<nbStacks; ++j)
+		stacks[j].resize(0);
+
+	// put all cells in the level range into the appropriate stacks
+	for (int y = 0; y < h; ++y)
+	{
+		for (int x = 0; x < w; ++x)
+		{
+			const rcCompactCell& c = chf.cells[x+y*w];
+			for (int i = (int)c.index, ni = (int)(c.index+c.count); i < ni; ++i)
+			{
+				if (chf.areas[i] == RC_NULL_AREA || srcReg[i] != 0)
+					continue;
+
+				int level = chf.dist[i] >> loglevelsPerStack;
+				int sId = startLevel - level;
+				if (sId >= (int)nbStacks)
+					continue;
+				if (sId < 0)
+					sId = 0;
+
+				stacks[sId].push(x);
+				stacks[sId].push(y);
+				stacks[sId].push(i);
+			}
+		}
+	}
+}
+
+
+static void appendStacks(rcIntArray& srcStack, rcIntArray& dstStack,
+						 unsigned short* srcReg)
+{
+	for (int j=0; j<srcStack.size(); j+=3)
+	{
+		int i = srcStack[j+2];
+		if ((i < 0) || (srcReg[i] != 0))
+			continue;
+		dstStack.push(srcStack[j]);
+		dstStack.push(srcStack[j+1]);
+		dstStack.push(srcStack[j+2]);
+	}
+}
+
 struct rcRegion
 {
 	inline rcRegion(unsigned short i) :
@@ -441,7 +517,11 @@ struct rcRegion
 		id(i),
 		areaType(0),
 		remap(false),
-		visited(false)
+		visited(false),
+		overlap(false),
+		connectsToBorder(false),
+		ymin(0xffff),
+		ymax(0)
 	{}
 	
 	int spanCount;					// Number of spans belonging to this region
@@ -449,6 +529,9 @@ struct rcRegion
 	unsigned char areaType;			// Are type.
 	bool remap;
 	bool visited;
+	bool overlap;
+	bool connectsToBorder;
+	unsigned short ymin, ymax;
 	rcIntArray connections;
 	rcIntArray floors;
 };
@@ -678,25 +761,26 @@ static void walkContour(int x, int y, int i, int dir,
 	// Remove adjacent duplicates.
 	if (cont.size() > 1)
 	{
-		for (int i = 0; i < cont.size(); )
+		for (int j = 0; j < cont.size(); )
 		{
-			int ni = (i+1) % cont.size();
-			if (cont[i] == cont[ni])
+			int nj = (j+1) % cont.size();
+			if (cont[j] == cont[nj])
 			{
-				for (int j = i; j < cont.size()-1; ++j)
-					cont[j] = cont[j+1];
+				for (int k = j; k < cont.size()-1; ++k)
+					cont[k] = cont[k+1];
 				cont.pop();
 			}
 			else
-				++i;
+				++j;
 		}
 	}
 }
 
-static bool filterSmallRegions(rcContext* ctx, int minRegionArea, int mergeRegionSize,
-							   unsigned short& maxRegionId,
-							   rcCompactHeightfield& chf,
-							   unsigned short* srcReg)
+
+static bool mergeAndFilterRegions(rcContext* ctx, int minRegionArea, int mergeRegionSize,
+								  unsigned short& maxRegionId,
+								  rcCompactHeightfield& chf,
+								  unsigned short* srcReg, rcIntArray& overlaps)
 {
 	const int w = chf.width;
 	const int h = chf.height;
@@ -705,7 +789,7 @@ static bool filterSmallRegions(rcContext* ctx, int minRegionArea, int mergeRegio
 	rcRegion* regions = (rcRegion*)rcAlloc(sizeof(rcRegion)*nreg, RC_ALLOC_TEMP);
 	if (!regions)
 	{
-		ctx->log(RC_LOG_ERROR, "filterSmallRegions: Out of memory 'regions' (%d).", nreg);
+		ctx->log(RC_LOG_ERROR, "mergeAndFilterRegions: Out of memory 'regions' (%d).", nreg);
 		return false;
 	}
 
@@ -728,7 +812,6 @@ static bool filterSmallRegions(rcContext* ctx, int minRegionArea, int mergeRegio
 				rcRegion& reg = regions[r];
 				reg.spanCount++;
 				
-				
 				// Update floors.
 				for (int j = (int)c.index; j < ni; ++j)
 				{
@@ -736,6 +819,8 @@ static bool filterSmallRegions(rcContext* ctx, int minRegionArea, int mergeRegio
 					unsigned short floorId = srcReg[j];
 					if (floorId == 0 || floorId >= nreg)
 						continue;
+					if (floorId == r)
+						reg.overlap = true;
 					addUniqueFloorRegion(reg, floorId);
 				}
 				
@@ -806,14 +891,14 @@ static bool filterSmallRegions(rcContext* ctx, int minRegionArea, int mergeRegio
 					connectsToBorder = true;
 					continue;
 				}
-				rcRegion& nreg = regions[creg.connections[j]];
-				if (nreg.visited)
+				rcRegion& neireg = regions[creg.connections[j]];
+				if (neireg.visited)
 					continue;
-				if (nreg.id == 0 || (nreg.id & RC_BORDER_REG))
+				if (neireg.id == 0 || (neireg.id & RC_BORDER_REG))
 					continue;
 				// Visit
-				stack.push(nreg.id);
-				nreg.visited = true;
+				stack.push(neireg.id);
+				neireg.visited = true;
 			}
 		}
 		
@@ -831,7 +916,7 @@ static bool filterSmallRegions(rcContext* ctx, int minRegionArea, int mergeRegio
 			}
 		}
 	}
-		
+	
 	// Merge too small regions to neighbour regions.
 	int mergeCount = 0 ;
 	do
@@ -841,7 +926,9 @@ static bool filterSmallRegions(rcContext* ctx, int minRegionArea, int mergeRegio
 		{
 			rcRegion& reg = regions[i];
 			if (reg.id == 0 || (reg.id & RC_BORDER_REG))
-				continue;                       
+				continue;
+			if (reg.overlap)
+				continue;
 			if (reg.spanCount == 0)
 				continue;
 			
@@ -858,7 +945,7 @@ static bool filterSmallRegions(rcContext* ctx, int minRegionArea, int mergeRegio
 			{
 				if (reg.connections[j] & RC_BORDER_REG) continue;
 				rcRegion& mreg = regions[reg.connections[j]];
-				if (mreg.id == 0 || (mreg.id & RC_BORDER_REG)) continue;
+				if (mreg.id == 0 || (mreg.id & RC_BORDER_REG) || mreg.overlap) continue;
 				if (mreg.spanCount < smallest &&
 					canMergeWithRegion(reg, mreg) &&
 					canMergeWithRegion(mreg, reg))
@@ -922,6 +1009,224 @@ static bool filterSmallRegions(rcContext* ctx, int minRegionArea, int mergeRegio
 	}
 	maxRegionId = regIdGen;
 	
+	// Remap regions.
+	for (int i = 0; i < chf.spanCount; ++i)
+	{
+		if ((srcReg[i] & RC_BORDER_REG) == 0)
+			srcReg[i] = regions[srcReg[i]].id;
+	}
+
+	// Return regions that we found to be overlapping.
+	for (int i = 0; i < nreg; ++i)
+		if (regions[i].overlap)
+			overlaps.push(regions[i].id);
+
+	for (int i = 0; i < nreg; ++i)
+		regions[i].~rcRegion();
+	rcFree(regions);
+	
+	
+	return true;
+}
+
+
+static void addUniqueConnection(rcRegion& reg, int n)
+{
+	for (int i = 0; i < reg.connections.size(); ++i)
+		if (reg.connections[i] == n)
+			return;
+	reg.connections.push(n);
+}
+
+static bool mergeAndFilterLayerRegions(rcContext* ctx, int minRegionArea,
+									   unsigned short& maxRegionId,
+									   rcCompactHeightfield& chf,
+									   unsigned short* srcReg, rcIntArray& /*overlaps*/)
+{
+	const int w = chf.width;
+	const int h = chf.height;
+	
+	const int nreg = maxRegionId+1;
+	rcRegion* regions = (rcRegion*)rcAlloc(sizeof(rcRegion)*nreg, RC_ALLOC_TEMP);
+	if (!regions)
+	{
+		ctx->log(RC_LOG_ERROR, "mergeAndFilterLayerRegions: Out of memory 'regions' (%d).", nreg);
+		return false;
+	}
+	
+	// Construct regions
+	for (int i = 0; i < nreg; ++i)
+		new(&regions[i]) rcRegion((unsigned short)i);
+	
+	// Find region neighbours and overlapping regions.
+	rcIntArray lregs(32);
+	for (int y = 0; y < h; ++y)
+	{
+		for (int x = 0; x < w; ++x)
+		{
+			const rcCompactCell& c = chf.cells[x+y*w];
+
+			lregs.resize(0);
+			
+			for (int i = (int)c.index, ni = (int)(c.index+c.count); i < ni; ++i)
+			{
+				const rcCompactSpan& s = chf.spans[i];
+				const unsigned short ri = srcReg[i];
+				if (ri == 0 || ri >= nreg) continue;
+				rcRegion& reg = regions[ri];
+				
+				reg.spanCount++;
+				
+				reg.ymin = rcMin(reg.ymin, s.y);
+				reg.ymax = rcMax(reg.ymax, s.y);
+				
+				// Collect all region layers.
+				lregs.push(ri);
+				
+				// Update neighbours
+				for (int dir = 0; dir < 4; ++dir)
+				{
+					if (rcGetCon(s, dir) != RC_NOT_CONNECTED)
+					{
+						const int ax = x + rcGetDirOffsetX(dir);
+						const int ay = y + rcGetDirOffsetY(dir);
+						const int ai = (int)chf.cells[ax+ay*w].index + rcGetCon(s, dir);
+						const unsigned short rai = srcReg[ai];
+						if (rai > 0 && rai < nreg && rai != ri)
+							addUniqueConnection(reg, rai);
+						if (rai & RC_BORDER_REG)
+							reg.connectsToBorder = true;
+					}
+				}
+				
+			}
+			
+			// Update overlapping regions.
+			for (int i = 0; i < lregs.size()-1; ++i)
+			{
+				for (int j = i+1; j < lregs.size(); ++j)
+				{
+					if (lregs[i] != lregs[j])
+					{
+						rcRegion& ri = regions[lregs[i]];
+						rcRegion& rj = regions[lregs[j]];
+						addUniqueFloorRegion(ri, lregs[j]);
+						addUniqueFloorRegion(rj, lregs[i]);
+					}
+				}
+			}
+			
+		}
+	}
+
+	// Create 2D layers from regions.
+	unsigned short layerId = 1;
+
+	for (int i = 0; i < nreg; ++i)
+		regions[i].id = 0;
+
+	// Merge montone regions to create non-overlapping areas.
+	rcIntArray stack(32);
+	for (int i = 1; i < nreg; ++i)
+	{
+		rcRegion& root = regions[i];
+		// Skip already visited.
+		if (root.id != 0)
+			continue;
+		
+		// Start search.
+		root.id = layerId;
+
+		stack.resize(0);
+		stack.push(i);
+		
+		while (stack.size() > 0)
+		{
+			// Pop front
+			rcRegion& reg = regions[stack[0]];
+			for (int j = 0; j < stack.size()-1; ++j)
+				stack[j] = stack[j+1];
+			stack.resize(stack.size()-1);
+			
+			const int ncons = (int)reg.connections.size();
+			for (int j = 0; j < ncons; ++j)
+			{
+				const int nei = reg.connections[j];
+				rcRegion& regn = regions[nei];
+				// Skip already visited.
+				if (regn.id != 0)
+					continue;
+				// Skip if the neighbour is overlapping root region.
+				bool overlap = false;
+				for (int k = 0; k < root.floors.size(); k++)
+				{
+					if (root.floors[k] == nei)
+					{
+						overlap = true;
+						break;
+					}
+				}
+				if (overlap)
+					continue;
+					
+				// Deepen
+				stack.push(nei);
+					
+				// Mark layer id
+				regn.id = layerId;
+				// Merge current layers to root.
+				for (int k = 0; k < regn.floors.size(); ++k)
+					addUniqueFloorRegion(root, regn.floors[k]);
+				root.ymin = rcMin(root.ymin, regn.ymin);
+				root.ymax = rcMax(root.ymax, regn.ymax);
+				root.spanCount += regn.spanCount;
+				regn.spanCount = 0;
+				root.connectsToBorder = root.connectsToBorder || regn.connectsToBorder;
+			}
+		}
+		
+		layerId++;
+	}
+	
+	// Remove small regions
+	for (int i = 0; i < nreg; ++i)
+	{
+		if (regions[i].spanCount > 0 && regions[i].spanCount < minRegionArea && !regions[i].connectsToBorder)
+		{
+			unsigned short reg = regions[i].id;
+			for (int j = 0; j < nreg; ++j)
+				if (regions[j].id == reg)
+					regions[j].id = 0;
+		}
+	}
+	
+	// Compress region Ids.
+	for (int i = 0; i < nreg; ++i)
+	{
+		regions[i].remap = false;
+		if (regions[i].id == 0) continue;				// Skip nil regions.
+		if (regions[i].id & RC_BORDER_REG) continue;    // Skip external regions.
+		regions[i].remap = true;
+	}
+	
+	unsigned short regIdGen = 0;
+	for (int i = 0; i < nreg; ++i)
+	{
+		if (!regions[i].remap)
+			continue;
+		unsigned short oldId = regions[i].id;
+		unsigned short newId = ++regIdGen;
+		for (int j = i; j < nreg; ++j)
+		{
+			if (regions[j].id == oldId)
+			{
+				regions[j].id = newId;
+				regions[j].remap = false;
+			}
+		}
+	}
+	maxRegionId = regIdGen;
+	
 	// Remap regions.
 	for (int i = 0; i < chf.spanCount; ++i)
 	{
@@ -936,6 +1241,8 @@ static bool filterSmallRegions(rcContext* ctx, int minRegionArea, int mergeRegio
 	return true;
 }
 
+
+
 /// @par
 /// 
 /// This is usually the second to the last step in creating a fully built
@@ -950,7 +1257,7 @@ bool rcBuildDistanceField(rcContext* ctx, rcCompactHeightfield& chf)
 {
 	rcAssert(ctx);
 	
-	ctx->startTimer(RC_TIMER_BUILD_DISTANCEFIELD);
+	rcScopedTimer timer(ctx, RC_TIMER_BUILD_DISTANCEFIELD);
 	
 	if (chf.dist)
 	{
@@ -974,25 +1281,23 @@ bool rcBuildDistanceField(rcContext* ctx, rcCompactHeightfield& chf)
 	
 	unsigned short maxDist = 0;
 
-	ctx->startTimer(RC_TIMER_BUILD_DISTANCEFIELD_DIST);
-	
-	calculateDistanceField(chf, src, maxDist);
-	chf.maxDistance = maxDist;
-	
-	ctx->stopTimer(RC_TIMER_BUILD_DISTANCEFIELD_DIST);
-	
-	ctx->startTimer(RC_TIMER_BUILD_DISTANCEFIELD_BLUR);
-	
-	// Blur
-	if (boxBlur(chf, 1, src, dst) != src)
-		rcSwap(src, dst);
-	
-	// Store distance.
-	chf.dist = src;
-	
-	ctx->stopTimer(RC_TIMER_BUILD_DISTANCEFIELD_BLUR);
+	{
+		rcScopedTimer timerDist(ctx, RC_TIMER_BUILD_DISTANCEFIELD_DIST);
 
-	ctx->stopTimer(RC_TIMER_BUILD_DISTANCEFIELD);
+		calculateDistanceField(chf, src, maxDist);
+		chf.maxDistance = maxDist;
+	}
+
+	{
+		rcScopedTimer timerBlur(ctx, RC_TIMER_BUILD_DISTANCEFIELD_BLUR);
+
+		// Blur
+		if (boxBlur(chf, 1, src, dst) != src)
+			rcSwap(src, dst);
+
+		// Store distance.
+		chf.dist = src;
+	}
 	
 	rcFree(dst);
 	
@@ -1052,13 +1357,13 @@ bool rcBuildRegionsMonotone(rcContext* ctx, rcCompactHeightfield& chf,
 {
 	rcAssert(ctx);
 	
-	ctx->startTimer(RC_TIMER_BUILD_REGIONS);
+	rcScopedTimer timer(ctx, RC_TIMER_BUILD_REGIONS);
 	
 	const int w = chf.width;
 	const int h = chf.height;
 	unsigned short id = 1;
 	
-	rcScopedDelete<unsigned short> srcReg = (unsigned short*)rcAlloc(sizeof(unsigned short)*chf.spanCount, RC_ALLOC_TEMP);
+	rcScopedDelete<unsigned short> srcReg((unsigned short*)rcAlloc(sizeof(unsigned short)*chf.spanCount, RC_ALLOC_TEMP));
 	if (!srcReg)
 	{
 		ctx->log(RC_LOG_ERROR, "rcBuildRegionsMonotone: Out of memory 'src' (%d).", chf.spanCount);
@@ -1067,7 +1372,7 @@ bool rcBuildRegionsMonotone(rcContext* ctx, rcCompactHeightfield& chf,
 	memset(srcReg,0,sizeof(unsigned short)*chf.spanCount);
 
 	const int nsweeps = rcMax(chf.width,chf.height);
-	rcScopedDelete<rcSweepSpan> sweeps = (rcSweepSpan*)rcAlloc(sizeof(rcSweepSpan)*nsweeps, RC_ALLOC_TEMP);
+	rcScopedDelete<rcSweepSpan> sweeps((rcSweepSpan*)rcAlloc(sizeof(rcSweepSpan)*nsweeps, RC_ALLOC_TEMP));
 	if (!sweeps)
 	{
 		ctx->log(RC_LOG_ERROR, "rcBuildRegionsMonotone: Out of memory 'sweeps' (%d).", nsweeps);
@@ -1181,20 +1486,22 @@ bool rcBuildRegionsMonotone(rcContext* ctx, rcCompactHeightfield& chf,
 		}
 	}
 
-	ctx->startTimer(RC_TIMER_BUILD_REGIONS_FILTER);
 
-	// Filter out small regions.
-	chf.maxRegions = id;
-	if (!filterSmallRegions(ctx, minRegionArea, mergeRegionArea, chf.maxRegions, chf, srcReg))
-		return false;
+	{
+		rcScopedTimer timerFilter(ctx, RC_TIMER_BUILD_REGIONS_FILTER);
 
-	ctx->stopTimer(RC_TIMER_BUILD_REGIONS_FILTER);
+		// Merge regions and filter out small regions.
+		rcIntArray overlaps;
+		chf.maxRegions = id;
+		if (!mergeAndFilterRegions(ctx, minRegionArea, mergeRegionArea, chf.maxRegions, chf, srcReg, overlaps))
+			return false;
+
+		// Monotone partitioning does not generate overlapping regions.
+	}
 	
 	// Store the result out.
 	for (int i = 0; i < chf.spanCount; ++i)
 		chf.spans[i].reg = srcReg[i];
-	
-	ctx->stopTimer(RC_TIMER_BUILD_REGIONS);
 
 	return true;
 }
@@ -1223,12 +1530,12 @@ bool rcBuildRegions(rcContext* ctx, rcCompactHeightfield& chf,
 {
 	rcAssert(ctx);
 	
-	ctx->startTimer(RC_TIMER_BUILD_REGIONS);
+	rcScopedTimer timer(ctx, RC_TIMER_BUILD_REGIONS);
 	
 	const int w = chf.width;
 	const int h = chf.height;
 	
-	rcScopedDelete<unsigned short> buf = (unsigned short*)rcAlloc(sizeof(unsigned short)*chf.spanCount*4, RC_ALLOC_TEMP);
+	rcScopedDelete<unsigned short> buf((unsigned short*)rcAlloc(sizeof(unsigned short)*chf.spanCount*4, RC_ALLOC_TEMP));
 	if (!buf)
 	{
 		ctx->log(RC_LOG_ERROR, "rcBuildRegions: Out of memory 'tmp' (%d).", chf.spanCount*4);
@@ -1236,7 +1543,13 @@ bool rcBuildRegions(rcContext* ctx, rcCompactHeightfield& chf,
 	}
 	
 	ctx->startTimer(RC_TIMER_BUILD_REGIONS_WATERSHED);
-	
+
+	const int LOG_NB_STACKS = 3;
+	const int NB_STACKS = 1 << LOG_NB_STACKS;
+	rcIntArray lvlStacks[NB_STACKS];
+	for (int i=0; i<NB_STACKS; ++i)
+		lvlStacks[i].resize(1024);
+
 	rcIntArray stack(1024);
 	rcIntArray visited(1024);
 	
@@ -1262,6 +1575,13 @@ bool rcBuildRegions(rcContext* ctx, rcCompactHeightfield& chf,
 		// Make sure border will not overflow.
 		const int bw = rcMin(w, borderSize);
 		const int bh = rcMin(h, borderSize);
+
+		if (regionId > 0xFFFB)
+		{
+			ctx->log(RC_LOG_ERROR, "rcBuildRegions: Region ID overflow");
+			return false;
+		}
+		
 		// Paint regions
 		paintRectRegion(0, bw, 0, h, regionId|RC_BORDER_REG, chf, srcReg); regionId++;
 		paintRectRegion(w-bw, w, 0, h, regionId|RC_BORDER_REG, chf, srcReg); regionId++;
@@ -1271,44 +1591,60 @@ bool rcBuildRegions(rcContext* ctx, rcCompactHeightfield& chf,
 		chf.borderSize = borderSize;
 	}
 	
+	int sId = -1;
 	while (level > 0)
 	{
 		level = level >= 2 ? level-2 : 0;
-		
-		ctx->startTimer(RC_TIMER_BUILD_REGIONS_EXPAND);
-		
-		// Expand current regions until no empty connected cells found.
-		if (expandRegions(expandIters, level, chf, srcReg, srcDist, dstReg, dstDist, stack) != srcReg)
+		sId = (sId+1) & (NB_STACKS-1);
+
+//		ctx->startTimer(RC_TIMER_DIVIDE_TO_LEVELS);
+
+		if (sId == 0)
+			sortCellsByLevel(level, chf, srcReg, NB_STACKS, lvlStacks, 1);
+		else 
+			appendStacks(lvlStacks[sId-1], lvlStacks[sId], srcReg); // copy left overs from last level
+
+//		ctx->stopTimer(RC_TIMER_DIVIDE_TO_LEVELS);
+
 		{
-			rcSwap(srcReg, dstReg);
-			rcSwap(srcDist, dstDist);
-		}
-		
-		ctx->stopTimer(RC_TIMER_BUILD_REGIONS_EXPAND);
-		
-		ctx->startTimer(RC_TIMER_BUILD_REGIONS_FLOOD);
-		
-		// Mark new regions with IDs.
-		for (int y = 0; y < h; ++y)
-		{
-			for (int x = 0; x < w; ++x)
+			rcScopedTimer timerExpand(ctx, RC_TIMER_BUILD_REGIONS_EXPAND);
+
+			// Expand current regions until no empty connected cells found.
+			if (expandRegions(expandIters, level, chf, srcReg, srcDist, dstReg, dstDist, lvlStacks[sId], false) != srcReg)
 			{
-				const rcCompactCell& c = chf.cells[x+y*w];
-				for (int i = (int)c.index, ni = (int)(c.index+c.count); i < ni; ++i)
-				{
-					if (chf.dist[i] < level || srcReg[i] != 0 || chf.areas[i] == RC_NULL_AREA)
-						continue;
-					if (floodRegion(x, y, i, level, regionId, chf, srcReg, srcDist, stack))
-						regionId++;
-				}
+				rcSwap(srcReg, dstReg);
+				rcSwap(srcDist, dstDist);
 			}
 		}
 		
-		ctx->stopTimer(RC_TIMER_BUILD_REGIONS_FLOOD);
+		{
+			rcScopedTimer timerFloor(ctx, RC_TIMER_BUILD_REGIONS_FLOOD);
+
+			// Mark new regions with IDs.
+			for (int j = 0; j<lvlStacks[sId].size(); j += 3)
+			{
+				int x = lvlStacks[sId][j];
+				int y = lvlStacks[sId][j+1];
+				int i = lvlStacks[sId][j+2];
+				if (i >= 0 && srcReg[i] == 0)
+				{
+					if (floodRegion(x, y, i, level, regionId, chf, srcReg, srcDist, stack))
+					{
+						if (regionId == 0xFFFF)
+						{
+							ctx->log(RC_LOG_ERROR, "rcBuildRegions: Region ID overflow");
+							return false;
+						}
+						
+						regionId++;
+					}
+				}
+			}
+		}
 	}
 	
 	// Expand current regions until no empty connected cells found.
-	if (expandRegions(expandIters*8, 0, chf, srcReg, srcDist, dstReg, dstDist, stack) != srcReg)
+	if (expandRegions(expandIters*8, 0, chf, srcReg, srcDist, dstReg, dstDist, stack, true) != srcReg)
 	{
 		rcSwap(srcReg, dstReg);
 		rcSwap(srcDist, dstDist);
@@ -1316,22 +1652,179 @@ bool rcBuildRegions(rcContext* ctx, rcCompactHeightfield& chf,
 	
 	ctx->stopTimer(RC_TIMER_BUILD_REGIONS_WATERSHED);
 	
-	ctx->startTimer(RC_TIMER_BUILD_REGIONS_FILTER);
-	
-	// Filter out small regions.
-	chf.maxRegions = regionId;
-	if (!filterSmallRegions(ctx, minRegionArea, mergeRegionArea, chf.maxRegions, chf, srcReg))
-		return false;
-	
-	ctx->stopTimer(RC_TIMER_BUILD_REGIONS_FILTER);
+	{
+		rcScopedTimer timerFilter(ctx, RC_TIMER_BUILD_REGIONS_FILTER);
+
+		// Merge regions and filter out smalle regions.
+		rcIntArray overlaps;
+		chf.maxRegions = regionId;
+		if (!mergeAndFilterRegions(ctx, minRegionArea, mergeRegionArea, chf.maxRegions, chf, srcReg, overlaps))
+			return false;
+
+		// If overlapping regions were found during merging, split those regions.
+		if (overlaps.size() > 0)
+		{
+			ctx->log(RC_LOG_ERROR, "rcBuildRegions: %d overlapping regions.", overlaps.size());
+		}
+	}
 		
 	// Write the result out.
 	for (int i = 0; i < chf.spanCount; ++i)
 		chf.spans[i].reg = srcReg[i];
 	
-	ctx->stopTimer(RC_TIMER_BUILD_REGIONS);
-	
 	return true;
 }
 
 
+bool rcBuildLayerRegions(rcContext* ctx, rcCompactHeightfield& chf,
+						 const int borderSize, const int minRegionArea)
+{
+	rcAssert(ctx);
+	
+	rcScopedTimer timer(ctx, RC_TIMER_BUILD_REGIONS);
+	
+	const int w = chf.width;
+	const int h = chf.height;
+	unsigned short id = 1;
+	
+	rcScopedDelete<unsigned short> srcReg((unsigned short*)rcAlloc(sizeof(unsigned short)*chf.spanCount, RC_ALLOC_TEMP));
+	if (!srcReg)
+	{
+		ctx->log(RC_LOG_ERROR, "rcBuildRegionsMonotone: Out of memory 'src' (%d).", chf.spanCount);
+		return false;
+	}
+	memset(srcReg,0,sizeof(unsigned short)*chf.spanCount);
+	
+	const int nsweeps = rcMax(chf.width,chf.height);
+	rcScopedDelete<rcSweepSpan> sweeps((rcSweepSpan*)rcAlloc(sizeof(rcSweepSpan)*nsweeps, RC_ALLOC_TEMP));
+	if (!sweeps)
+	{
+		ctx->log(RC_LOG_ERROR, "rcBuildRegionsMonotone: Out of memory 'sweeps' (%d).", nsweeps);
+		return false;
+	}
+	
+	
+	// Mark border regions.
+	if (borderSize > 0)
+	{
+		// Make sure border will not overflow.
+		const int bw = rcMin(w, borderSize);
+		const int bh = rcMin(h, borderSize);
+		// Paint regions
+		paintRectRegion(0, bw, 0, h, id|RC_BORDER_REG, chf, srcReg); id++;
+		paintRectRegion(w-bw, w, 0, h, id|RC_BORDER_REG, chf, srcReg); id++;
+		paintRectRegion(0, w, 0, bh, id|RC_BORDER_REG, chf, srcReg); id++;
+		paintRectRegion(0, w, h-bh, h, id|RC_BORDER_REG, chf, srcReg); id++;
+		
+		chf.borderSize = borderSize;
+	}
+	
+	rcIntArray prev(256);
+	
+	// Sweep one line at a time.
+	for (int y = borderSize; y < h-borderSize; ++y)
+	{
+		// Collect spans from this row.
+		prev.resize(id+1);
+		memset(&prev[0],0,sizeof(int)*id);
+		unsigned short rid = 1;
+		
+		for (int x = borderSize; x < w-borderSize; ++x)
+		{
+			const rcCompactCell& c = chf.cells[x+y*w];
+			
+			for (int i = (int)c.index, ni = (int)(c.index+c.count); i < ni; ++i)
+			{
+				const rcCompactSpan& s = chf.spans[i];
+				if (chf.areas[i] == RC_NULL_AREA) continue;
+				
+				// -x
+				unsigned short previd = 0;
+				if (rcGetCon(s, 0) != RC_NOT_CONNECTED)
+				{
+					const int ax = x + rcGetDirOffsetX(0);
+					const int ay = y + rcGetDirOffsetY(0);
+					const int ai = (int)chf.cells[ax+ay*w].index + rcGetCon(s, 0);
+					if ((srcReg[ai] & RC_BORDER_REG) == 0 && chf.areas[i] == chf.areas[ai])
+						previd = srcReg[ai];
+				}
+				
+				if (!previd)
+				{
+					previd = rid++;
+					sweeps[previd].rid = previd;
+					sweeps[previd].ns = 0;
+					sweeps[previd].nei = 0;
+				}
+				
+				// -y
+				if (rcGetCon(s,3) != RC_NOT_CONNECTED)
+				{
+					const int ax = x + rcGetDirOffsetX(3);
+					const int ay = y + rcGetDirOffsetY(3);
+					const int ai = (int)chf.cells[ax+ay*w].index + rcGetCon(s, 3);
+					if (srcReg[ai] && (srcReg[ai] & RC_BORDER_REG) == 0 && chf.areas[i] == chf.areas[ai])
+					{
+						unsigned short nr = srcReg[ai];
+						if (!sweeps[previd].nei || sweeps[previd].nei == nr)
+						{
+							sweeps[previd].nei = nr;
+							sweeps[previd].ns++;
+							prev[nr]++;
+						}
+						else
+						{
+							sweeps[previd].nei = RC_NULL_NEI;
+						}
+					}
+				}
+				
+				srcReg[i] = previd;
+			}
+		}
+		
+		// Create unique ID.
+		for (int i = 1; i < rid; ++i)
+		{
+			if (sweeps[i].nei != RC_NULL_NEI && sweeps[i].nei != 0 &&
+				prev[sweeps[i].nei] == (int)sweeps[i].ns)
+			{
+				sweeps[i].id = sweeps[i].nei;
+			}
+			else
+			{
+				sweeps[i].id = id++;
+			}
+		}
+		
+		// Remap IDs
+		for (int x = borderSize; x < w-borderSize; ++x)
+		{
+			const rcCompactCell& c = chf.cells[x+y*w];
+			
+			for (int i = (int)c.index, ni = (int)(c.index+c.count); i < ni; ++i)
+			{
+				if (srcReg[i] > 0 && srcReg[i] < rid)
+					srcReg[i] = sweeps[srcReg[i]].id;
+			}
+		}
+	}
+	
+	
+	{
+		rcScopedTimer timerFilter(ctx, RC_TIMER_BUILD_REGIONS_FILTER);
+
+		// Merge monotone regions to layers and remove small regions.
+		rcIntArray overlaps;
+		chf.maxRegions = id;
+		if (!mergeAndFilterLayerRegions(ctx, minRegionArea, chf.maxRegions, chf, srcReg, overlaps))
+			return false;
+	}
+	
+	
+	// Store the result out.
+	for (int i = 0; i < chf.spanCount; ++i)
+		chf.spans[i].reg = srcReg[i];
+	
+	return true;
+}

extern/recastnavigation/Recast/Source/RecastTimer.cpp

@@ -1,58 +0,0 @@
-#include "RecastTimer.h"
-
-#if defined(WIN32)
-
-// Win32
-#include <windows.h>
-
-rcTimeVal rcGetPerformanceTimer()
-{
-	__int64 count;
-	QueryPerformanceCounter((LARGE_INTEGER*)&count);
-	return count;
-}
-
-int rcGetDeltaTimeUsec(rcTimeVal start, rcTimeVal end)
-{
-	static __int64 freq = 0;
-	if (freq == 0)
-		QueryPerformanceFrequency((LARGE_INTEGER*)&freq);
-	__int64 elapsed = end - start;
-	return (int)(elapsed*1000000 / freq);
-}
-
-#elif defined(__MACH__)
-
-// OSX
-#include <mach/mach_time.h>
-
-rcTimeVal rcGetPerformanceTimer()
-{
-	return mach_absolute_time();
-}
-
-int rcGetDeltaTimeUsec(rcTimeVal start, rcTimeVal end)
-{
-	static mach_timebase_info_data_t timebaseInfo;
-	if (timebaseInfo.denom == 0)
-		mach_timebase_info(&timebaseInfo);
-	uint64_t elapsed = end - start;
-	uint64_t nanosec = elapsed * timebaseInfo.numer / timebaseInfo.denom;
-	return (int)(nanosec / 1000);
-}
-
-#else
-
-// TODO: Linux, etc
-
-rcTimeVal rcGetPerformanceTimer()
-{
-	return 0;
-}
-
-int rcGetDeltaTimeUsec(rcTimeVal start, rcTimeVal end)
-{
-	return 0;
-}
-
-#endif

extern/recastnavigation/readme-blender.txt

@@ -0,0 +1,22 @@
+The version of Recast is 1.5.0, from:
+https://github.com/recastnavigation/recastnavigation
+Changes made:
+  * Recast/Source/RecastMesh.cpp: made buildMeshAdjacency() non-static so it can be used with recast-capi
+  * Recast/Include/Recast.h: Added forward declaration for buildMeshAdjacency()
+
+The following additional files were added:
+  * recast-capi.cpp
+  * recast-capi.h
+These expose a C interface to the Recast library, which has only C++ headers.
+
+The version of Detour is 1.4, from:
+https://code.google.com/archive/p/recastnavigation/downloads
+Changes made:
+  * DetourStatNavMesh.h: use more portable definition of DT_STAT_NAVMESH_MAGIC
+  * DetourStatNavMesh.cpp: comment out some unused variables to avoid compiler warnings
+  * DetourStatNavMeshBuilder.h: add forward declaration for createBVTree
+  * DetourStatNavMeshBuilder.cpp: made createBVTree non-static for use with recast-capi
+
+The CMakeLists.txt file has been added, since the original software does not include build files for the libraries.
+
+~rdb

extern/recastnavigation/recast-capi.cpp

@@ -68,17 +68,41 @@ int recast_createHeightfield(struct recast_heightfield *hf, int width, int heigh
 }
 
 void recast_markWalkableTriangles(const float walkableSlopeAngle,const float *verts, int nv,
-			const int *tris, int nt, unsigned char *flags)
+			const int *tris, int nt, unsigned char *areas)
 {
 	INIT_SCTX();
-	rcMarkWalkableTriangles(sctx, walkableSlopeAngle, verts, nv, tris, nt, flags);
+	rcMarkWalkableTriangles(sctx, walkableSlopeAngle, verts, nv, tris, nt, areas);
 }
 
-void recast_rasterizeTriangles(const float *verts, int nv, const int *tris,
-			const unsigned char *flags, int nt, struct recast_heightfield *solid)
+void recast_clearUnwalkableTriangles(const float walkableSlopeAngle, const float* verts, int nv,
+			const int* tris, int nt, unsigned char* areas)
 {
 	INIT_SCTX();
-	rcRasterizeTriangles(sctx, verts, nv, tris, flags, nt, *(rcHeightfield *) solid);
+	rcClearUnwalkableTriangles(sctx, walkableSlopeAngle, verts, nv, tris, nt, areas);
+}
+
+int recast_addSpan(struct recast_heightfield *hf, const int x, const int y,
+			const unsigned short smin, const unsigned short smax,
+			const unsigned char area, const int flagMergeThr)
+{
+	INIT_SCTX();
+	return rcAddSpan(sctx, *(rcHeightfield *) hf, x, y, smin, smax, area, flagMergeThr);
+}
+
+int recast_rasterizeTriangle(const float *v0, const float *v1, const float *v2,
+			const unsigned char area, struct recast_heightfield *solid,
+			const int flagMergeThr)
+{
+	INIT_SCTX();
+	return rcRasterizeTriangle(sctx, v0, v1, v2, area, *(rcHeightfield *) solid, flagMergeThr);
+}
+
+int recast_rasterizeTriangles(const float *verts, const int nv, const int *tris,
+			const unsigned char *areas, const int nt, struct recast_heightfield *solid,
+			const int flagMergeThr)
+{
+	INIT_SCTX();
+	return rcRasterizeTriangles(sctx, verts, nv, tris, areas, nt, *(rcHeightfield *) solid, flagMergeThr);
 }
 
 void recast_filterLedgeSpans(const int walkableHeight, const int walkableClimb,
@@ -100,6 +124,22 @@ void recast_filterLowHangingWalkableObstacles(const int walkableClimb, struct re
 	rcFilterLowHangingWalkableObstacles(sctx, walkableClimb, *(rcHeightfield *) solid);
 }
 
+int recast_getHeightFieldSpanCount(struct recast_heightfield *hf)
+{
+	INIT_SCTX();
+	return rcGetHeightFieldSpanCount(sctx, *(rcHeightfield *) hf);
+}
+
+struct recast_heightfieldLayerSet *recast_newHeightfieldLayerSet(void)
+{
+	return (struct recast_heightfieldLayerSet *) rcAllocHeightfieldLayerSet();
+}
+
+void recast_destroyHeightfieldLayerSet(struct recast_heightfieldLayerSet *lset)
+{
+	rcFreeHeightfieldLayerSet( (rcHeightfieldLayerSet *) lset);
+}
+
 struct recast_compactHeightfield *recast_newCompactHeightfield(void)
 {
 	return (struct recast_compactHeightfield *) rcAllocCompactHeightfield();
@@ -124,18 +164,68 @@ int recast_erodeWalkableArea(int radius, struct recast_compactHeightfield *chf)
 	return rcErodeWalkableArea(sctx, radius, *(rcCompactHeightfield *) chf);
 }
 
+int recast_medianFilterWalkableArea(struct recast_compactHeightfield *chf)
+{
+	INIT_SCTX();
+	return rcMedianFilterWalkableArea(sctx, *(rcCompactHeightfield *) chf);
+}
+
+void recast_markBoxArea(const float *bmin, const float *bmax, unsigned char areaId,
+			struct recast_compactHeightfield *chf)
+{
+	INIT_SCTX();
+	rcMarkBoxArea(sctx, bmin, bmax, areaId, *(rcCompactHeightfield *) chf);
+}
+
+void recast_markConvexPolyArea(const float* verts, const int nverts,
+			const float hmin, const float hmax, unsigned char areaId,
+			struct recast_compactHeightfield *chf)
+{
+	INIT_SCTX();
+	rcMarkConvexPolyArea(sctx, verts, nverts, hmin, hmax, areaId, *(rcCompactHeightfield *) chf);
+}
+
+int recast_offsetPoly(const float* verts, const int nverts,
+			const float offset, float *outVerts, const int maxOutVerts)
+{
+	return rcOffsetPoly(verts, nverts, offset, outVerts, maxOutVerts);
+}
+
+void recast_markCylinderArea(const float* pos, const float r, const float h,
+			unsigned char areaId, struct recast_compactHeightfield *chf)
+{
+	INIT_SCTX();
+	rcMarkCylinderArea(sctx, pos, r, h, areaId, *(rcCompactHeightfield *) chf);
+}
+
 int recast_buildDistanceField(struct recast_compactHeightfield *chf)
 {
 	INIT_SCTX();
 	return rcBuildDistanceField(sctx, *(rcCompactHeightfield *) chf);
 }
 
-int recast_buildRegions(struct recast_compactHeightfield *chf, int borderSize,
-	int minRegionSize, int mergeRegionSize)
+int recast_buildRegions(struct recast_compactHeightfield *chf,
+			const int borderSize, const int minRegionArea, const int mergeRegionArea)
 {
 	INIT_SCTX();
 	return rcBuildRegions(sctx, *(rcCompactHeightfield *) chf, borderSize,
-				minRegionSize, mergeRegionSize);
+				minRegionArea, mergeRegionArea);
+}
+
+int recast_buildLayerRegions(struct recast_compactHeightfield *chf,
+			const int borderSize, const int minRegionArea)
+{
+	INIT_SCTX();
+	return rcBuildLayerRegions(sctx, *(rcCompactHeightfield *) chf, borderSize,
+				minRegionArea);
+}
+
+int recast_buildRegionsMonotone(struct recast_compactHeightfield *chf,
+			const int borderSize, const int minRegionArea, const int mergeRegionArea)
+{
+	INIT_SCTX();
+	return rcBuildRegionsMonotone(sctx, *(rcCompactHeightfield *) chf, borderSize,
+				minRegionArea, mergeRegionArea);
 }
 
 struct recast_contourSet *recast_newContourSet(void)
@@ -149,10 +239,11 @@ void recast_destroyContourSet(struct recast_contourSet *contourSet)
 }
 
 int recast_buildContours(struct recast_compactHeightfield *chf,
-			const float maxError, const int maxEdgeLen, struct recast_contourSet *cset)
+			const float maxError, const int maxEdgeLen, struct recast_contourSet *cset,
+			const int buildFlags)
 {
 	INIT_SCTX();
-	return rcBuildContours(sctx, *(rcCompactHeightfield *) chf, maxError, maxEdgeLen, *(rcContourSet *) cset);
+	return rcBuildContours(sctx, *(rcCompactHeightfield *) chf, maxError, maxEdgeLen, *(rcContourSet *) cset, buildFlags);
 }
 
 struct recast_polyMesh *recast_newPolyMesh(void)
@@ -165,10 +256,22 @@ void recast_destroyPolyMesh(struct recast_polyMesh *polyMesh)
 	rcFreePolyMesh((rcPolyMesh *) polyMesh);
 }
 
-int recast_buildPolyMesh(struct recast_contourSet *cset, int nvp, struct recast_polyMesh *mesh)
+int recast_buildPolyMesh(struct recast_contourSet *cset, const int nvp, struct recast_polyMesh *mesh)
 {
 	INIT_SCTX();
-	return rcBuildPolyMesh(sctx, *(rcContourSet *) cset, nvp, * (rcPolyMesh *) mesh);
+	return rcBuildPolyMesh(sctx, *(rcContourSet *) cset, nvp, *(rcPolyMesh *) mesh);
+}
+
+int recast_mergePolyMeshes(struct recast_polyMesh **meshes, const int nmeshes, struct recast_polyMesh *mesh)
+{
+	INIT_SCTX();
+	return rcMergePolyMeshes(sctx, (rcPolyMesh **) meshes, nmeshes, *(rcPolyMesh *) mesh);
+}
+
+int recast_copyPolyMesh(const struct recast_polyMesh *src, struct recast_polyMesh *dst)
+{
+	INIT_SCTX();
+	return rcCopyPolyMesh(sctx, *(const rcPolyMesh *) src, *(rcPolyMesh *) dst);
 }
 
 unsigned short *recast_polyMeshGetVerts(struct recast_polyMesh *mesh, int *nverts)
@@ -240,6 +343,12 @@ int recast_buildPolyMeshDetail(const struct recast_polyMesh *mesh, const struct
 			sampleDist, sampleMaxError, *(rcPolyMeshDetail *) dmesh);
 }
 
+int recast_mergePolyMeshDetails(struct recast_polyMeshDetail **meshes, const int nmeshes, struct recast_polyMeshDetail *mesh)
+{
+	INIT_SCTX();
+	return rcMergePolyMeshDetails(sctx, (rcPolyMeshDetail **) meshes, nmeshes, *(rcPolyMeshDetail *) mesh);
+}
+
 float *recast_polyMeshDetailGetVerts(struct recast_polyMeshDetail *mesh, int *nverts)
 {
 	rcPolyMeshDetail *dmesh = (rcPolyMeshDetail *)mesh;

extern/recastnavigation/recast-capi.h

@@ -38,12 +38,13 @@ struct recast_polyMesh;
 struct recast_polyMeshDetail;
 struct recast_heightfield;
 struct recast_compactHeightfield;
+struct recast_heightfieldLayerSet;
 struct recast_contourSet;
 
-enum recast_SpanFlags
+enum recast_BuildContoursFlags
 {
-	RECAST_WALKABLE = 0x01,
-	RECAST_REACHABLE = 0x02
+	RECAST_CONTOUR_TESS_WALL_EDGES = 0x01,
+	RECAST_CONTOUR_TESS_AREA_EDGES = 0x02,
 };
 
 int recast_buildMeshAdjacency(unsigned short* polys, const int npolys,
@@ -61,10 +62,22 @@ int recast_createHeightfield(struct recast_heightfield *hf, int width, int heigh
 			const float *bmin, const float* bmax, float cs, float ch);
 
 void recast_markWalkableTriangles(const float walkableSlopeAngle,const float *verts, int nv,
-			const int *tris, int nt, unsigned char *flags);
+			const int *tris, int nt, unsigned char *areas);
 
-void recast_rasterizeTriangles(const float *verts, int nv, const int *tris,
-			const unsigned char *flags, int nt, struct recast_heightfield *solid);
+void recast_clearUnwalkableTriangles(const float walkableSlopeAngle, const float* verts, int nv,
+			const int* tris, int nt, unsigned char* areas);
+
+int recast_addSpan(struct recast_heightfield *hf, const int x, const int y,
+			const unsigned short smin, const unsigned short smax,
+			const unsigned char area, const int flagMergeThr);
+
+int recast_rasterizeTriangle(const float* v0, const float* v1, const float* v2,
+			const unsigned char area, struct recast_heightfield *solid,
+			const int flagMergeThr);
+
+int recast_rasterizeTriangles(const float *verts, const int nv, const int *tris,
+			const unsigned char *areas, const int nt, struct recast_heightfield *solid,
+			const int flagMergeThr);
 
 void recast_filterLedgeSpans(const int walkableHeight, const int walkableClimb,
 			struct recast_heightfield *solid);
@@ -73,6 +86,12 @@ void recast_filterWalkableLowHeightSpans(int walkableHeight, struct recast_heigh
 
 void recast_filterLowHangingWalkableObstacles(const int walkableClimb, struct recast_heightfield *solid);
 
+int recast_getHeightFieldSpanCount(struct recast_heightfield *hf);
+
+struct recast_heightfieldLayerSet *recast_newHeightfieldLayerSet(void);
+
+void recast_destroyHeightfieldLayerSet(struct recast_heightfieldLayerSet *lset);
+
 struct recast_compactHeightfield *recast_newCompactHeightfield(void);
 
 void recast_destroyCompactHeightfield(struct recast_compactHeightfield *compactHeightfield);
@@ -82,10 +101,31 @@ int recast_buildCompactHeightfield(const int walkableHeight, const int walkableC
 
 int recast_erodeWalkableArea(int radius, struct recast_compactHeightfield *chf);
 
+int recast_medianFilterWalkableArea(struct recast_compactHeightfield *chf);
+
+void recast_markBoxArea(const float *bmin, const float *bmax, unsigned char areaId,
+			struct recast_compactHeightfield *chf);
+
+void recast_markConvexPolyArea(const float* verts, const int nverts,
+			const float hmin, const float hmax, unsigned char areaId,
+			struct recast_compactHeightfield *chf);
+
+int recast_offsetPoly(const float* verts, const int nverts,
+			const float offset, float *outVerts, const int maxOutVerts);
+
+void recast_markCylinderArea(const float* pos, const float r, const float h,
+			unsigned char areaId, struct recast_compactHeightfield *chf);
+
 int recast_buildDistanceField(struct recast_compactHeightfield *chf);
 
-int recast_buildRegions(struct recast_compactHeightfield *chf, int borderSize,
-	int minRegionSize, int mergeRegionSize);
+int recast_buildRegions(struct recast_compactHeightfield *chf,
+			const int borderSize, const int minRegionArea, const int mergeRegionArea);
+
+int recast_buildLayerRegions(struct recast_compactHeightfield *chf,
+			const int borderSize, const int minRegionArea);
+
+int recast_buildRegionsMonotone(struct recast_compactHeightfield *chf,
+			const int borderSize, const int minRegionArea, const int mergeRegionArea);
 
 /* Contour set */
 
@@ -94,7 +134,8 @@ struct recast_contourSet *recast_newContourSet(void);
 void recast_destroyContourSet(struct recast_contourSet *contourSet);
 
 int recast_buildContours(struct recast_compactHeightfield *chf,
-			const float maxError, const int maxEdgeLen, struct recast_contourSet *cset);
+			const float maxError, const int maxEdgeLen, struct recast_contourSet *cset,
+			const int buildFlags);
 
 /* Poly mesh */
 
@@ -102,7 +143,11 @@ struct recast_polyMesh *recast_newPolyMesh(void);
 
 void recast_destroyPolyMesh(struct recast_polyMesh *polyMesh);
 
-int recast_buildPolyMesh(struct recast_contourSet *cset, int nvp, struct recast_polyMesh *mesh);
+int recast_buildPolyMesh(struct recast_contourSet *cset, const int nvp, struct recast_polyMesh *mesh);
+
+int recast_mergePolyMeshes(struct recast_polyMesh **meshes, const int nmeshes, struct recast_polyMesh *mesh);
+
+int recast_copyPolyMesh(const struct recast_polyMesh *src, struct recast_polyMesh *dst);
 
 unsigned short *recast_polyMeshGetVerts(struct recast_polyMesh *mesh, int *nverts);
 
@@ -121,6 +166,8 @@ void recast_destroyPolyMeshDetail(struct recast_polyMeshDetail *polyMeshDetail);
 int recast_buildPolyMeshDetail(const struct recast_polyMesh *mesh, const struct recast_compactHeightfield *chf,
 			const float sampleDist, const float sampleMaxError, struct recast_polyMeshDetail *dmesh);
 
+int recast_mergePolyMeshDetails(struct recast_polyMeshDetail **meshes, const int nmeshes, struct recast_polyMeshDetail *mesh);
+
 float *recast_polyMeshDetailGetVerts(struct recast_polyMeshDetail *mesh, int *nverts);
 
 unsigned char *recast_polyMeshDetailGetTris(struct recast_polyMeshDetail *mesh, int *ntris);

extern/sdlew/auto/sdlew_gen.sh

@@ -1,250 +0,0 @@
-#!/bin/bash
-
-SDL="SDL2"
-INCLUDE_DIR="/usr/include/${SDL}"
-SCRIPT=`realpath -s $0`
-DIR=`dirname $SCRIPT`
-DIR=`dirname $DIR`
-
-mkdir -p $DIR/include/${SDL}
-mkdir -p $DIR/src
-
-rm -rf $DIR/include/${SDL}/*.h
-rm -rf $DIR/src/sdlew.c
-
-echo "Generating sdlew headers..."
-
-UNSUPPORTED="SDL_MemoryBarrierRelease SDL_MemoryBarrierAcquire SDL_AtomicCAS SDL_AtomicCASPtr \
-  SDL_iPhoneSetAnimationCallback SDL_iPhoneSetEventPump SDL_AndroidGetJNIEnv SDL_AndroidGetActivity \
-  SDL_AndroidGetActivity SDL_AndroidGetInternalStoragePath SDL_AndroidGetExternalStorageState \
-  SDL_AndroidGetExternalStoragePath SDL_CreateShapedWindow SDL_IsShapedWindow tSDL_SetWindowShape \
-  SDL_GetShapedWindowMode"
-
-for header in $INCLUDE_DIR/*; do
-  filename=`basename $header`
-  cat $header \
-    | sed -r 's/extern DECLSPEC ((const )?[a-z0-9_]+(\s\*)?)\s?SDLCALL /typedef \1 SDLCALL t/i' \
-    > $DIR/include/${SDL}/$filename
-
-  line_num=`cat $DIR/include/${SDL}/$filename | grep -n "Ends C function" | cut -d : -f 1`
-  if [ ! -z "$line_num" ]; then
-    functions=`grep -E 'typedef [A-Za-z0-9_ \*]+ SDLCALL' $DIR/include/${SDL}/$filename \
-      | sed -r 's/typedef [A-Za-z0-9_ \*]+ SDLCALL t([a-z0-9_]+).*/extern t\1 *\1;/i'`
-    functions=`echo "${functions}" | sed -e 's/[\/&]/\\\&/g'`
-    echo "$functions" | while read function; do
-      if [ -z "$function" ]; then
-        continue;
-      fi
-      func_name=`echo $function | cut -d '*' -f 2 | sed -r 's/;//'`
-      if [ ! -z "`echo "$UNSUPPORTED" | grep $func_name`" ]; then
-        continue;
-      fi
-      if [ "$func_name" == "SDL_memcpy" ]; then
-        line_num=`cat $DIR/include/${SDL}/$filename | grep -n "SDL_memcpy4" | cut -d : -f 1`
-        sed -ri "${line_num}s/(.*)/${function}\n\1/" $DIR/include/${SDL}/$filename
-      else
-        sed -ri "${line_num}s/(.*)/${function}\n\1/" $DIR/include/${SDL}/$filename
-      fi
-      line_num=`cat $DIR/include/${SDL}/$filename | grep -n "Ends C function" | cut -d : -f 1`
-    done
-    line_num=`cat $DIR/include/${SDL}/$filename | grep -n "Ends C function" | cut -d : -f 1`
-    sed -ri "${line_num}s/(.*)/\n\1/" $DIR/include/${SDL}/$filename
-  fi
-
-  if [ $filename == "SDL_stdinc.h"  ]; then
-    cat $header | grep -E '#if(def)? (defined\()?HAVE_' | sed -r 's/#if(def)? //' | while read check; do
-      func_names=`cat $DIR/include/${SDL}/$filename \
-                   | grep -A 8 "$check\$" \
-                   | grep -v struct \
-                   | grep 'typedef' \
-                   | sed -r 's/typedef [a-z0-9_ \*]+ SDLCALL ([a-z0-9_]+).*/\1/i'`
-      full_check=`echo "${check}" | sed -e 's/[\/&]/\\\&/g'`
-      if [ ! -z "`echo $full_check | grep defined`"  ]; then
-        full_check="#if !($full_check)"
-      else
-        full_check="#ifndef $full_check"
-      fi
-      for func_name in $func_names; do
-         line_num=`grep -n "extern ${func_name} \*" $DIR/include/${SDL}/$filename | cut -d : -f 1`
-         let prev_num=line_num-1
-         if [ -z "`cat $DIR/include/${SDL}/$filename | head -n $prev_num | tail -n 1 | grep '#if'`" ]; then
-           sed -ri "${line_num}s/(.*)/$full_check \/* GEN_CHECK_MARKER *\/\n\1\n#endif \/* GEN_CHECK_MARKER *\//" $DIR/include/${SDL}/$filename
-         fi
-      done
-    done
-  fi
-done
-
-cat << EOF > $DIR/include/sdlew.h
-/*
- * Copyright 2014 Blender Foundation
- *
- * Licensed under the Apache License, Version 2.0 (the "License");
- * you may not use this file except in compliance with the License.
- * You may obtain a copy of the License at
- *
- * http://www.apache.org/licenses/LICENSE-2.0
- *
- * Unless required by applicable law or agreed to in writing, software
- * distributed under the License is distributed on an "AS IS" BASIS,
- * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
- * See the License for the specific language governing permissions and
- * limitations under the License
- */
-
-#ifndef __SDL_EW_H__
-#define __SDL_EW_H__
-
-#ifdef __cplusplus
-extern "C" {
-#endif
-
-enum {
-  SDLEW_SUCCESS = 0,
-  SDLEW_ERROR_OPEN_FAILED = -1,
-  SDLEW_ERROR_ATEXIT_FAILED = -2,
-  SDLEW_ERROR_VERSION = -3,
-};
-
-int sdlewInit(void);
-
-#ifdef __cplusplus
-}
-#endif
-
-#endif  /* __SDL_EW_H__ */
-EOF
-
-echo "Generating sdlew sources..."
-
-cat << EOF > $DIR/src/sdlew.c
-/*
- * Copyright 2014 Blender Foundation
- *
- * Licensed under the Apache License, Version 2.0 (the "License");
- * you may not use this file except in compliance with the License.
- * You may obtain a copy of the License at
- *
- * http://www.apache.org/licenses/LICENSE-2.0
- *
- * Unless required by applicable law or agreed to in writing, software
- * distributed under the License is distributed on an "AS IS" BASIS,
- * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
- * See the License for the specific language governing permissions and
- * limitations under the License
- */
-
-#ifdef _MSC_VER
-#  define snprintf _snprintf
-#  define popen _popen
-#  define pclose _pclose
-#  define _CRT_SECURE_NO_WARNINGS
-#endif
-
-#include "sdlew.h"
-
-#include "${SDL}/SDL.h"
-#include "${SDL}/SDL_syswm.h"
-
-#include <assert.h>
-#include <stdio.h>
-#include <string.h>
-#include <sys/stat.h>
-
-#ifdef _WIN32
-#  define WIN32_LEAN_AND_MEAN
-#  define VC_EXTRALEAN
-#  include <windows.h>
-
-/* Utility macros. */
-
-typedef HMODULE DynamicLibrary;
-
-#  define dynamic_library_open(path)         LoadLibrary(path)
-#  define dynamic_library_close(lib)         FreeLibrary(lib)
-#  define dynamic_library_find(lib, symbol)  GetProcAddress(lib, symbol)
-#else
-#  include <dlfcn.h>
-
-typedef void* DynamicLibrary;
-
-#  define dynamic_library_open(path)         dlopen(path, RTLD_NOW)
-#  define dynamic_library_close(lib)         dlclose(lib)
-#  define dynamic_library_find(lib, symbol)  dlsym(lib, symbol)
-#endif
-
-#define SDL_LIBRARY_FIND_CHECKED(name) \
-        name = (t##name *)dynamic_library_find(lib, #name); \
-        assert(name);
-
-#define SDL_LIBRARY_FIND(name) \
-        name = (t##name *)dynamic_library_find(lib, #name);
-
-static DynamicLibrary lib;
-
-EOF
-
-content=`grep --no-filename -ER "extern tSDL|GEN_CHECK_MARKER" $DIR/include/${SDL}/`
-
-echo "$content" | sed -r 's/extern t([a-z0-9_]+).*/t\1 *\1;/gi' >> $DIR/src/sdlew.c
-
-cat << EOF >> $DIR/src/sdlew.c
-
-static void sdlewExit(void) {
-  if(lib != NULL) {
-    /*  Ignore errors. */
-    dynamic_library_close(lib);
-    lib = NULL;
-  }
-}
-
-/* Implementation function. */
-int sdlewInit(void) {
-  /* Library paths. */
-#ifdef _WIN32
-  /* Expected in c:/windows/system or similar, no path needed. */
-  const char *path = "SDL2.dll";
-#elif defined(__APPLE__)
-  /* Default installation path. */
-  const char *path = "/usr/local/cuda/lib/libSDL2.dylib";
-#else
-  const char *path = "libSDL2.so";
-#endif
-  static int initialized = 0;
-  static int result = 0;
-  int error;
-
-  if (initialized) {
-    return result;
-  }
-
-  initialized = 1;
-
-  error = atexit(sdlewExit);
-  if (error) {
-    result = SDLEW_ERROR_ATEXIT_FAILED;
-    return result;
-  }
-
-  /* Load library. */
-  lib = dynamic_library_open(path);
-
-  if (lib == NULL) {
-    result = SDLEW_ERROR_OPEN_FAILED;
-    return result;
-  }
-
-EOF
-
-echo "$content" | sed -r 's/extern t([a-z0-9_]+).*/  SDL_LIBRARY_FIND(\1);/gi' >> $DIR/src/sdlew.c
-
-cat << EOF >> $DIR/src/sdlew.c
-
-  result = SDLEW_SUCCESS;
-
-  return result;
-}
-EOF
-
-sed -i 's/\s\/\* GEN_CHECK_MARKER \*\///g' $DIR/src/sdlew.c
-sed -i 's/\s\/\* GEN_CHECK_MARKER \*\///g' $DIR/include/${SDL}/SDL_stdinc.h

extern/sdlew/auto/strip_comments.sh

@@ -1,15 +0,0 @@
-#!/bin/bash
-
-SDL="SDL2"
-INCLUDE_DIR="/usr/include/${SDL}"
-SCRIPT=`realpath -s $0`
-DIR=`dirname $SCRIPT`
-DIR=`dirname $DIR`
-
-for f in $DIR/include/${SDL}/*.h; do
-  file_name=`basename $f`
-  echo "Striping $file_name..."
-  sed -r ':a; s%(.*)/\*.*\*/%\1%; ta; /\/\*/ !b; N; ba' -i $f
-  sed 's/[ \t]*$//' -i $f
-  sed '/^$/N;/^\n$/D' -i $f
-done

intern/atomic/atomic_ops.h

@@ -1,11 +1,11 @@
 /*
- * Adopted from jemalloc with this license:
+ * Original code from jemalloc with this license:
  *
  * Copyright (C) 2002-2013 Jason Evans <jasone@canonware.com>.
  * All rights reserved.
  * Copyright (C) 2007-2012 Mozilla Foundation.  All rights reserved.
  * Copyright (C) 2009-2013 Facebook, Inc.  All rights reserved.
-
+ *
  * Redistribution and use in source and binary forms, with or without
  * modification, are permitted provided that the following conditions are met:
  * 1. Redistributions of source code must retain the above copyright notice(s),
@@ -13,7 +13,7 @@
  * 2. Redistributions in binary form must reproduce the above copyright notice(s),
  *    this list of conditions and the following disclaimer in the documentation
  *    and/or other materials provided with the distribution.
-
+ *
  * THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDER(S) ``AS IS'' AND ANY EXPRESS
  * OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE IMPLIED WARRANTIES OF
  * MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE ARE DISCLAIMED.  IN NO
@@ -24,63 +24,59 @@
  * LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT (INCLUDING NEGLIGENCE
  * OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE OF THIS SOFTWARE, EVEN IF
  * ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
+ *
+ * ***** 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.
+ *
+ * The Original Code is: adapted from jemalloc.
+ *
+ * ***** END GPL LICENSE BLOCK *****
+ */
+
+/**
+ * \file atomic_ops.h
+ * \ingroup Atomic
+ *
+ * \author Copyright (C) 2016 Blender Foundation, adapted from jemalloc.
+ * \brief Provides wrapper around system-specific atomic primitives, and some extensions (faked-atomic operations
+ *        over float numbers).
  */
 
 #ifndef __ATOMIC_OPS_H__
 #define __ATOMIC_OPS_H__
 
-#include <assert.h>
-
-#if defined (__APPLE__)
-#  include <libkern/OSAtomic.h>
-#elif defined(_MSC_VER)
-#  define NOGDI
-#  ifndef NOMINMAX
-#    define NOMINMAX
-#  endif
-#  define WIN32_LEAN_AND_MEAN
-#  include <windows.h>
-#elif defined(__arm__)
+#if defined(__arm__)
 /* Attempt to fix compilation error on Debian armel kernel.
  * arm7 architecture does have both 32 and 64bit atomics, however
  * it's gcc doesn't have __GCC_HAVE_SYNC_COMPARE_AND_SWAP_n defined.
  */
-#  define JE_FORCE_SYNC_COMPARE_AND_SWAP_8
+#  define JE_FORCE_SYNC_COMPARE_AND_SWAP_1
 #  define JE_FORCE_SYNC_COMPARE_AND_SWAP_4
+#  define JE_FORCE_SYNC_COMPARE_AND_SWAP_8
 #endif
 
-/* needed for int types */
-#include "../../source/blender/blenlib/BLI_sys_types.h"
-#include <stdlib.h>
-#include <stddef.h>
+#include "intern/atomic_ops_utils.h"
 
-/* little macro so inline keyword works */
-#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
-#endif
-
-/* This is becoming a bit nastier that it was originally foreseen,
- * consider using autoconfig detection instead.
- */
-#if defined(_M_X64) || defined(__amd64__) || defined(__x86_64__) || defined(__s390x__) || defined(__powerpc64__) || defined(__aarch64__) || (defined(__sparc__) && defined(__arch64__)) || defined(__alpha__) || defined(__mips64)
-#  define LG_SIZEOF_PTR 3
-#  define LG_SIZEOF_INT 2
-#else
-#  define LG_SIZEOF_PTR 2
-#  define LG_SIZEOF_INT 2
-#endif
-
-/************************/
+/******************************************************************************/
 /* Function prototypes. */
 
-#if (LG_SIZEOF_PTR == 3 || LG_SIZEOF_INT == 3)
+#if (LG_SIZEOF_PTR == 8 || LG_SIZEOF_INT == 8)
 ATOMIC_INLINE uint64_t atomic_add_uint64(uint64_t *p, uint64_t x);
 ATOMIC_INLINE uint64_t atomic_sub_uint64(uint64_t *p, uint64_t x);
 ATOMIC_INLINE uint64_t atomic_cas_uint64(uint64_t *v, uint64_t old, uint64_t _new);
@@ -90,6 +86,9 @@ ATOMIC_INLINE uint32_t atomic_add_uint32(uint32_t *p, uint32_t x);
 ATOMIC_INLINE uint32_t atomic_sub_uint32(uint32_t *p, uint32_t x);
 ATOMIC_INLINE uint32_t atomic_cas_uint32(uint32_t *v, uint32_t old, uint32_t _new);
 
+ATOMIC_INLINE uint32_t atomic_fetch_and_add_uint32(uint32_t *p, uint32_t x);
+
+ATOMIC_INLINE uint8_t atomic_fetch_and_or_uint8(uint8_t *p, uint8_t b);
 ATOMIC_INLINE uint8_t atomic_fetch_and_and_uint8(uint8_t *p, uint8_t b);
 
 ATOMIC_INLINE size_t atomic_add_z(size_t *p, size_t x);
@@ -100,395 +99,22 @@ ATOMIC_INLINE unsigned atomic_add_u(unsigned *p, unsigned x);
 ATOMIC_INLINE unsigned atomic_sub_u(unsigned *p, unsigned x);
 ATOMIC_INLINE unsigned atomic_cas_u(unsigned *v, unsigned old, unsigned _new);
 
-/******************************************************************************/
-/* 64-bit operations. */
-#if (LG_SIZEOF_PTR == 3 || LG_SIZEOF_INT == 3)
-#  ifdef __GCC_HAVE_SYNC_COMPARE_AND_SWAP_8
-ATOMIC_INLINE uint64_t
-atomic_add_uint64(uint64_t *p, uint64_t x)
-{
-	return __sync_add_and_fetch(p, x);
-}
-
-ATOMIC_INLINE uint64_t
-atomic_sub_uint64(uint64_t *p, uint64_t x)
-{
-	return __sync_sub_and_fetch(p, x);
-}
-
-ATOMIC_INLINE uint64_t
-atomic_cas_uint64(uint64_t *v, uint64_t old, uint64_t _new)
-{
-	return __sync_val_compare_and_swap(v, old, _new);
-}
-#elif (defined(_MSC_VER))
-ATOMIC_INLINE uint64_t
-atomic_add_uint64(uint64_t *p, uint64_t x)
-{
-	return InterlockedExchangeAdd64((int64_t *)p, (int64_t)x) + x;
-}
-
-ATOMIC_INLINE uint64_t
-atomic_sub_uint64(uint64_t *p, uint64_t x)
-{
-	return InterlockedExchangeAdd64((int64_t *)p, -((int64_t)x)) - x;
-}
-
-ATOMIC_INLINE uint64_t
-atomic_cas_uint64(uint64_t *v, uint64_t old, uint64_t _new)
-{
-	return InterlockedCompareExchange64((int64_t *)v, _new, old);
-}
-#elif (defined(__APPLE__))
-ATOMIC_INLINE uint64_t
-atomic_add_uint64(uint64_t *p, uint64_t x)
-{
-	return (uint64_t)OSAtomicAdd64((int64_t)x, (int64_t *)p);
-}
-
-ATOMIC_INLINE uint64_t
-atomic_sub_uint64(uint64_t *p, uint64_t x)
-{
-	return (uint64_t)OSAtomicAdd64(-((int64_t)x), (int64_t *)p);
-}
-
-ATOMIC_INLINE uint64_t
-atomic_cas_uint64(uint64_t *v, uint64_t old, uint64_t _new)
-{
-	uint64_t init_val = *v;
-	OSAtomicCompareAndSwap64((int64_t)old, (int64_t)_new, (int64_t *)v);
-	return init_val;
-}
-#  elif (defined(__amd64__) || defined(__x86_64__))
-ATOMIC_INLINE uint64_t
-atomic_add_uint64(uint64_t *p, uint64_t x)
-{
-	asm volatile (
-	    "lock; xaddq %0, %1;"
-	    : "+r" (x), "=m" (*p) /* Outputs. */
-	    : "m" (*p) /* Inputs. */
-	    );
-	return x;
-}
-
-ATOMIC_INLINE uint64_t
-atomic_sub_uint64(uint64_t *p, uint64_t x)
-{
-	x = (uint64_t)(-(int64_t)x);
-	asm volatile (
-	    "lock; xaddq %0, %1;"
-	    : "+r" (x), "=m" (*p) /* Outputs. */
-	    : "m" (*p) /* Inputs. */
-	    );
-	return x;
-}
-
-ATOMIC_INLINE uint64_t
-atomic_cas_uint64(uint64_t *v, uint64_t old, uint64_t _new)
-{
-	uint64_t ret;
-	asm volatile (
-	    "lock; cmpxchgq %2,%1"
-	    : "=a" (ret), "+m" (*v)
-	    : "r" (_new), "0" (old)
-	    : "memory");
-	return ret;
-}
-
-#  elif (defined(JEMALLOC_ATOMIC9))
-ATOMIC_INLINE uint64_t
-atomic_add_uint64(uint64_t *p, uint64_t x)
-{
-	/*
-	 * atomic_fetchadd_64() doesn't exist, but we only ever use this
-	 * function on LP64 systems, so atomic_fetchadd_long() will do.
-	 */
-	assert(sizeof(uint64_t) == sizeof(unsigned long));
-
-	return atomic_fetchadd_long(p, (unsigned long)x) + x;
-}
-
-ATOMIC_INLINE uint64_t
-atomic_sub_uint64(uint64_t *p, uint64_t x)
-{
-	assert(sizeof(uint64_t) == sizeof(unsigned long));
-
-	return atomic_fetchadd_long(p, (unsigned long)(-(long)x)) - x;
-}
-
-ATOMIC_INLINE uint64_t
-atomic_cas_uint64(uint64_t *v, uint64_t old, uint64_t _new)
-{
-	assert(sizeof(uint64_t) == sizeof(unsigned long));
-
-	return atomic_cmpset_long(v, old, _new);
-}
-#  elif (defined(JE_FORCE_SYNC_COMPARE_AND_SWAP_8))
-ATOMIC_INLINE uint64_t
-atomic_add_uint64(uint64_t *p, uint64_t x)
-{
-	return __sync_add_and_fetch(p, x);
-}
-
-ATOMIC_INLINE uint64_t
-atomic_sub_uint64(uint64_t *p, uint64_t x)
-{
-	return __sync_sub_and_fetch(p, x);
-}
-
-ATOMIC_INLINE uint64_t
-atomic_cas_uint64(uint64_t *v, uint64_t old, uint64_t _new)
-{
-	return __sync_val_compare_and_swap(v, old, _new);
-}
-#  else
-#    error "Missing implementation for 64-bit atomic operations"
-#  endif
-#endif
+/* WARNING! Float 'atomics' are really faked ones, those are actually closer to some kind of spinlock-sync'ed operation,
+ *          which means they are only efficient if collisions are highly unlikely (i.e. if probability of two threads
+ *          working on the same pointer at the same time is very low). */
+ATOMIC_INLINE float atomic_add_fl(float *p, const float x);
 
 /******************************************************************************/
-/* 32-bit operations. */
-#ifdef __GCC_HAVE_SYNC_COMPARE_AND_SWAP_4
-ATOMIC_INLINE uint32_t
-atomic_add_uint32(uint32_t *p, uint32_t x)
-{
-	return __sync_add_and_fetch(p, x);
-}
+/* Include system-dependent implementations. */
 
-ATOMIC_INLINE uint32_t
-atomic_sub_uint32(uint32_t *p, uint32_t x)
-{
-	return __sync_sub_and_fetch(p, x);
-}
-
-ATOMIC_INLINE uint32_t
-atomic_cas_uint32(uint32_t *v, uint32_t old, uint32_t _new)
-{
-   return __sync_val_compare_and_swap(v, old, _new);
-}
-#elif (defined(_MSC_VER))
-ATOMIC_INLINE uint32_t
-atomic_add_uint32(uint32_t *p, uint32_t x)
-{
-	return InterlockedExchangeAdd(p, x) + x;
-}
-
-ATOMIC_INLINE uint32_t
-atomic_sub_uint32(uint32_t *p, uint32_t x)
-{
-	return InterlockedExchangeAdd(p, -((int32_t)x)) - x;
-}
-
-ATOMIC_INLINE uint32_t
-atomic_cas_uint32(uint32_t *v, uint32_t old, uint32_t _new)
-{
-	return InterlockedCompareExchange((long *)v, _new, old);
-}
-#elif (defined(__APPLE__))
-ATOMIC_INLINE uint32_t
-atomic_add_uint32(uint32_t *p, uint32_t x)
-{
-	return (uint32_t)OSAtomicAdd32((int32_t)x, (int32_t *)p);
-}
-
-ATOMIC_INLINE uint32_t
-atomic_sub_uint32(uint32_t *p, uint32_t x)
-{
-	return (uint32_t)OSAtomicAdd32(-((int32_t)x), (int32_t *)p);
-}
-
-ATOMIC_INLINE uint32_t
-atomic_cas_uint32(uint32_t *v, uint32_t old, uint32_t _new)
-{
-	uint32_t init_val = *v;
-	OSAtomicCompareAndSwap32((int32_t)old, (int32_t)_new, (int32_t *)v);
-	return init_val;
-}
-#elif (defined(__i386__) || defined(__amd64__) || defined(__x86_64__))
-ATOMIC_INLINE uint32_t
-atomic_add_uint32(uint32_t *p, uint32_t x)
-{
-	asm volatile (
-	    "lock; xaddl %0, %1;"
-	    : "+r" (x), "=m" (*p) /* Outputs. */
-	    : "m" (*p) /* Inputs. */
-	    );
-	return x;
-}
-
-ATOMIC_INLINE uint32_t
-atomic_sub_uint32(uint32_t *p, uint32_t x)
-{
-	x = (uint32_t)(-(int32_t)x);
-	asm volatile (
-	    "lock; xaddl %0, %1;"
-	    : "+r" (x), "=m" (*p) /* Outputs. */
-	    : "m" (*p) /* Inputs. */
-	    );
-	return x;
-}
-
-ATOMIC_INLINE uint32_t
-atomic_cas_uint32(uint32_t *v, uint32_t old, uint32_t _new)
-{
-	uint32_t ret;
-	asm volatile (
-	    "lock; cmpxchgl %2,%1"
-	    : "=a" (ret), "+m" (*v)
-	    : "r" (_new), "0" (old)
-	    : "memory");
-	return ret;
-}
-#elif (defined(JEMALLOC_ATOMIC9))
-ATOMIC_INLINE uint32_t
-atomic_add_uint32(uint32_t *p, uint32_t x)
-{
-	return atomic_fetchadd_32(p, x) + x;
-}
-
-ATOMIC_INLINE uint32_t
-atomic_sub_uint32(uint32_t *p, uint32_t x)
-{
-	return atomic_fetchadd_32(p, (uint32_t)(-(int32_t)x)) - x;
-}
-
-ATOMIC_INLINE uint32_t
-atomic_cas_uint32(uint32_t *v, uint32_t old, uint32_t _new)
-{
-	return atomic_cmpset_32(v, old, _new);
-}
-#elif defined(JE_FORCE_SYNC_COMPARE_AND_SWAP_4)
-ATOMIC_INLINE uint32_t
-atomic_add_uint32(uint32_t *p, uint32_t x)
-{
-	return __sync_add_and_fetch(p, x);
-}
-
-ATOMIC_INLINE uint32_t
-atomic_sub_uint32(uint32_t *p, uint32_t x)
-{
-	return __sync_sub_and_fetch(p, x);
-}
-
-ATOMIC_INLINE uint32_t
-atomic_cas_uint32(uint32_t *v, uint32_t old, uint32_t _new)
-{
-	return __sync_val_compare_and_swap(v, old, _new);
-}
+/* Note that we are using _unix flavor as fallback here (it will raise precompiler errors as needed). */
+#if defined(_MSC_VER)
+#  include "intern/atomic_ops_msvc.h"
 #else
-#  error "Missing implementation for 32-bit atomic operations"
+#  include "intern/atomic_ops_unix.h"
 #endif
 
-/******************************************************************************/
-/* 8-bit operations. */
-#ifdef __GCC_HAVE_SYNC_COMPARE_AND_SWAP_1
-ATOMIC_INLINE uint8_t
-atomic_fetch_and_and_uint8(uint8_t *p, uint8_t b)
-{
-	return __sync_fetch_and_and(p, b);
-}
-#elif (defined(_MSC_VER))
-#include <intrin.h>
-#pragma intrinsic(_InterlockedAnd8)
-ATOMIC_INLINE uint8_t
-atomic_fetch_and_and_uint8(uint8_t *p, uint8_t b)
-{
-#if (LG_SIZEOF_PTR == 3 || LG_SIZEOF_INT == 3)
-	return InterlockedAnd8((char *)p, (char)b);
-#else
-	return _InterlockedAnd8((char *)p, (char)b);
-#endif
-}
-#else
-#  error "Missing implementation for 8-bit atomic operations"
-#endif
-
-/******************************************************************************/
-/* size_t operations. */
-ATOMIC_INLINE size_t
-atomic_add_z(size_t *p, size_t x)
-{
-	assert(sizeof(size_t) == 1 << LG_SIZEOF_PTR);
-
-#if (LG_SIZEOF_PTR == 3)
-	return (size_t)atomic_add_uint64((uint64_t *)p, (uint64_t)x);
-#elif (LG_SIZEOF_PTR == 2)
-	return (size_t)atomic_add_uint32((uint32_t *)p, (uint32_t)x);
-#endif
-}
-
-ATOMIC_INLINE size_t
-atomic_sub_z(size_t *p, size_t x)
-{
-	assert(sizeof(size_t) == 1 << LG_SIZEOF_PTR);
-
-#if (LG_SIZEOF_PTR == 3)
-	return (size_t)atomic_add_uint64((uint64_t *)p,
-	                                 (uint64_t)-((int64_t)x));
-#elif (LG_SIZEOF_PTR == 2)
-	return (size_t)atomic_add_uint32((uint32_t *)p,
-	                                 (uint32_t)-((int32_t)x));
-#endif
-}
-
-ATOMIC_INLINE size_t
-atomic_cas_z(size_t *v, size_t old, size_t _new)
-{
-	assert(sizeof(size_t) == 1 << LG_SIZEOF_PTR);
-
-#if (LG_SIZEOF_PTR == 3)
-	return (size_t)atomic_cas_uint64((uint64_t *)v,
-	                                 (uint64_t)old,
-	                                 (uint64_t)_new);
-#elif (LG_SIZEOF_PTR == 2)
-	return (size_t)atomic_cas_uint32((uint32_t *)v,
-	                                 (uint32_t)old,
-	                                 (uint32_t)_new);
-#endif
-}
-
-/******************************************************************************/
-/* unsigned operations. */
-ATOMIC_INLINE unsigned
-atomic_add_u(unsigned *p, unsigned x)
-{
-	assert(sizeof(unsigned) == 1 << LG_SIZEOF_INT);
-
-#if (LG_SIZEOF_INT == 3)
-	return (unsigned)atomic_add_uint64((uint64_t *)p, (uint64_t)x);
-#elif (LG_SIZEOF_INT == 2)
-	return (unsigned)atomic_add_uint32((uint32_t *)p, (uint32_t)x);
-#endif
-}
-
-ATOMIC_INLINE unsigned
-atomic_sub_u(unsigned *p, unsigned x)
-{
-	assert(sizeof(unsigned) == 1 << LG_SIZEOF_INT);
-
-#if (LG_SIZEOF_INT == 3)
-	return (unsigned)atomic_add_uint64((uint64_t *)p,
-	                                   (uint64_t)-((int64_t)x));
-#elif (LG_SIZEOF_INT == 2)
-	return (unsigned)atomic_add_uint32((uint32_t *)p,
-	                                   (uint32_t)-((int32_t)x));
-#endif
-}
-
-ATOMIC_INLINE unsigned
-atomic_cas_u(unsigned *v, unsigned old, unsigned _new)
-{
-	assert(sizeof(unsigned) == 1 << LG_SIZEOF_INT);
-
-#if (LG_SIZEOF_PTR == 3)
-	return (unsigned)atomic_cas_uint64((uint64_t *)v,
-	                                   (uint64_t)old,
-	                                   (uint64_t)_new);
-#elif (LG_SIZEOF_PTR == 2)
-	return (unsigned)atomic_cas_uint32((uint32_t *)v,
-	                                   (uint32_t)old,
-	                                   (uint32_t)_new);
-#endif
-}
+/* Include 'fake' atomic extensions, built over real atomic primitives. */
+#include "intern/atomic_ops_ext.h"
 
 #endif /* __ATOMIC_OPS_H__ */

intern/atomic/intern/atomic_ops_ext.h

@@ -0,0 +1,146 @@
+/*
+ * Original code from jemalloc with this license:
+ *
+ * Copyright (C) 2002-2013 Jason Evans <jasone@canonware.com>.
+ * All rights reserved.
+ * Copyright (C) 2007-2012 Mozilla Foundation.  All rights reserved.
+ * Copyright (C) 2009-2013 Facebook, Inc.  All rights reserved.
+ *
+ * Redistribution and use in source and binary forms, with or without
+ * modification, are permitted provided that the following conditions are met:
+ * 1. Redistributions of source code must retain the above copyright notice(s),
+ *    this list of conditions and the following disclaimer.
+ * 2. Redistributions in binary form must reproduce the above copyright notice(s),
+ *    this list of conditions and the following disclaimer in the documentation
+ *    and/or other materials provided with the distribution.
+ *
+ * THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDER(S) ``AS IS'' AND ANY EXPRESS
+ * OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE IMPLIED WARRANTIES OF
+ * MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE ARE DISCLAIMED.  IN NO
+ * EVENT SHALL THE COPYRIGHT HOLDER(S) BE LIABLE FOR ANY DIRECT, INDIRECT,
+ * INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT
+ * LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, DATA, OR
+ * PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY THEORY OF
+ * LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT (INCLUDING NEGLIGENCE
+ * OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE OF THIS SOFTWARE, EVEN IF
+ * ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
+ *
+ * ***** 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.
+ *
+ * The Original Code is: adapted from jemalloc.
+ *
+ * ***** END GPL LICENSE BLOCK *****
+ */
+
+#ifndef __ATOMIC_OPS_EXT_H__
+#define __ATOMIC_OPS_EXT_H__
+
+#include "atomic_ops_utils.h"
+
+/******************************************************************************/
+/* size_t operations. */
+ATOMIC_INLINE size_t atomic_add_z(size_t *p, size_t x)
+{
+	assert(sizeof(size_t) == LG_SIZEOF_PTR);
+
+#if (LG_SIZEOF_PTR == 8)
+	return (size_t)atomic_add_uint64((uint64_t *)p, (uint64_t)x);
+#elif (LG_SIZEOF_PTR == 4)
+	return (size_t)atomic_add_uint32((uint32_t *)p, (uint32_t)x);
+#endif
+}
+
+ATOMIC_INLINE size_t atomic_sub_z(size_t *p, size_t x)
+{
+	assert(sizeof(size_t) == LG_SIZEOF_PTR);
+
+#if (LG_SIZEOF_PTR == 8)
+	return (size_t)atomic_add_uint64((uint64_t *)p, (uint64_t)-((int64_t)x));
+#elif (LG_SIZEOF_PTR == 4)
+	return (size_t)atomic_add_uint32((uint32_t *)p, (uint32_t)-((int32_t)x));
+#endif
+}
+
+ATOMIC_INLINE size_t atomic_cas_z(size_t *v, size_t old, size_t _new)
+{
+	assert(sizeof(size_t) == LG_SIZEOF_PTR);
+
+#if (LG_SIZEOF_PTR == 8)
+	return (size_t)atomic_cas_uint64((uint64_t *)v, (uint64_t)old, (uint64_t)_new);
+#elif (LG_SIZEOF_PTR == 4)
+	return (size_t)atomic_cas_uint32((uint32_t *)v, (uint32_t)old, (uint32_t)_new);
+#endif
+}
+
+/******************************************************************************/
+/* unsigned operations. */
+ATOMIC_INLINE unsigned atomic_add_u(unsigned *p, unsigned x)
+{
+	assert(sizeof(unsigned) == LG_SIZEOF_INT);
+
+#if (LG_SIZEOF_INT == 8)
+	return (unsigned)atomic_add_uint64((uint64_t *)p, (uint64_t)x);
+#elif (LG_SIZEOF_INT == 4)
+	return (unsigned)atomic_add_uint32((uint32_t *)p, (uint32_t)x);
+#endif
+}
+
+ATOMIC_INLINE unsigned atomic_sub_u(unsigned *p, unsigned x)
+{
+	assert(sizeof(unsigned) == LG_SIZEOF_INT);
+
+#if (LG_SIZEOF_INT == 8)
+	return (unsigned)atomic_add_uint64((uint64_t *)p, (uint64_t)-((int64_t)x));
+#elif (LG_SIZEOF_INT == 4)
+	return (unsigned)atomic_add_uint32((uint32_t *)p, (uint32_t)-((int32_t)x));
+#endif
+}
+
+ATOMIC_INLINE unsigned atomic_cas_u(unsigned *v, unsigned old, unsigned _new)
+{
+	assert(sizeof(unsigned) == LG_SIZEOF_INT);
+
+#if (LG_SIZEOF_INT == 8)
+	return (unsigned)atomic_cas_uint64((uint64_t *)v, (uint64_t)old, (uint64_t)_new);
+#elif (LG_SIZEOF_INT == 4)
+	return (unsigned)atomic_cas_uint32((uint32_t *)v, (uint32_t)old, (uint32_t)_new);
+#endif
+}
+
+/******************************************************************************/
+/* float operations. */
+
+ATOMIC_INLINE float atomic_add_fl(float *p, const float x)
+{
+	assert(sizeof(float) == sizeof(uint32_t));
+
+	float oldval, newval;
+	uint32_t prevval;
+
+	do {  /* Note that since collisions are unlikely, loop will nearly always run once. */
+		oldval = *p;
+		newval = oldval + x;
+		prevval = atomic_cas_uint32((uint32_t *)p, *(uint32_t *)(&oldval), *(uint32_t *)(&newval));
+	} while (UNLIKELY(prevval != *(uint32_t *)(&oldval)));
+
+	return newval;
+}
+
+#endif /* __ATOMIC_OPS_EXT_H__ */

intern/atomic/intern/atomic_ops_msvc.h

@@ -0,0 +1,107 @@
+/*
+ * Adopted from jemalloc with this license:
+ *
+ * Copyright (C) 2002-2013 Jason Evans <jasone@canonware.com>.
+ * All rights reserved.
+ * Copyright (C) 2007-2012 Mozilla Foundation.  All rights reserved.
+ * Copyright (C) 2009-2013 Facebook, Inc.  All rights reserved.
+
+ * Redistribution and use in source and binary forms, with or without
+ * modification, are permitted provided that the following conditions are met:
+ * 1. Redistributions of source code must retain the above copyright notice(s),
+ *    this list of conditions and the following disclaimer.
+ * 2. Redistributions in binary form must reproduce the above copyright notice(s),
+ *    this list of conditions and the following disclaimer in the documentation
+ *    and/or other materials provided with the distribution.
+
+ * THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDER(S) ``AS IS'' AND ANY EXPRESS
+ * OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE IMPLIED WARRANTIES OF
+ * MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE ARE DISCLAIMED.  IN NO
+ * EVENT SHALL THE COPYRIGHT HOLDER(S) BE LIABLE FOR ANY DIRECT, INDIRECT,
+ * INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT
+ * LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, DATA, OR
+ * PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY THEORY OF
+ * LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT (INCLUDING NEGLIGENCE
+ * OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE OF THIS SOFTWARE, EVEN IF
+ * ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
+ */
+
+#ifndef __ATOMIC_OPS_MSVC_H__
+#define __ATOMIC_OPS_MSVC_H__
+
+#include "atomic_ops_utils.h"
+
+#define NOGDI
+#ifndef NOMINMAX
+#  define NOMINMAX
+#endif
+#define WIN32_LEAN_AND_MEAN
+
+#include <windows.h>
+#include <intrin.h>
+
+/******************************************************************************/
+/* 64-bit operations. */
+#if (LG_SIZEOF_PTR == 8 || LG_SIZEOF_INT == 8)
+ATOMIC_INLINE uint64_t atomic_add_uint64(uint64_t *p, uint64_t x)
+{
+	return InterlockedExchangeAdd64((int64_t *)p, (int64_t)x) + x;
+}
+
+ATOMIC_INLINE uint64_t atomic_sub_uint64(uint64_t *p, uint64_t x)
+{
+	return InterlockedExchangeAdd64((int64_t *)p, -((int64_t)x)) - x;
+}
+
+ATOMIC_INLINE uint64_t atomic_cas_uint64(uint64_t *v, uint64_t old, uint64_t _new)
+{
+	return InterlockedCompareExchange64((int64_t *)v, _new, old);
+}
+#endif
+
+/******************************************************************************/
+/* 32-bit operations. */
+ATOMIC_INLINE uint32_t atomic_add_uint32(uint32_t *p, uint32_t x)
+{
+	return InterlockedExchangeAdd(p, x) + x;
+}
+
+ATOMIC_INLINE uint32_t atomic_sub_uint32(uint32_t *p, uint32_t x)
+{
+	return InterlockedExchangeAdd(p, -((int32_t)x)) - x;
+}
+
+ATOMIC_INLINE uint32_t atomic_cas_uint32(uint32_t *v, uint32_t old, uint32_t _new)
+{
+	return InterlockedCompareExchange((long *)v, _new, old);
+}
+
+ATOMIC_INLINE uint32_t atomic_fetch_and_add_uint32(uint32_t *p, uint32_t x)
+{
+	return InterlockedExchangeAdd(p, x);
+}
+
+/******************************************************************************/
+/* 8-bit operations. */
+
+#pragma intrinsic(_InterlockedAnd8)
+ATOMIC_INLINE uint8_t atomic_fetch_and_and_uint8(uint8_t *p, uint8_t b)
+{
+#if (LG_SIZEOF_PTR == 8 || LG_SIZEOF_INT == 8)
+	return InterlockedAnd8((char *)p, (char)b);
+#else
+	return _InterlockedAnd8((char *)p, (char)b);
+#endif
+}
+
+#pragma intrinsic(_InterlockedOr8)
+ATOMIC_INLINE uint8_t atomic_fetch_and_or_uint8(uint8_t *p, uint8_t b)
+{
+#if (LG_SIZEOF_PTR == 8 || LG_SIZEOF_INT == 8)
+	return InterlockedOr8((char *)p, (char)b);
+#else
+	return _InterlockedOr8((char *)p, (char)b);
+#endif
+}
+
+#endif /* __ATOMIC_OPS_MSVC_H__ */

intern/atomic/intern/atomic_ops_unix.h

@@ -0,0 +1,190 @@
+/*
+ * Original code from jemalloc with this license:
+ *
+ * Copyright (C) 2002-2013 Jason Evans <jasone@canonware.com>.
+ * All rights reserved.
+ * Copyright (C) 2007-2012 Mozilla Foundation.  All rights reserved.
+ * Copyright (C) 2009-2013 Facebook, Inc.  All rights reserved.
+ *
+ * Redistribution and use in source and binary forms, with or without
+ * modification, are permitted provided that the following conditions are met:
+ * 1. Redistributions of source code must retain the above copyright notice(s),
+ *    this list of conditions and the following disclaimer.
+ * 2. Redistributions in binary form must reproduce the above copyright notice(s),
+ *    this list of conditions and the following disclaimer in the documentation
+ *    and/or other materials provided with the distribution.
+ *
+ * THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDER(S) ``AS IS'' AND ANY EXPRESS
+ * OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE IMPLIED WARRANTIES OF
+ * MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE ARE DISCLAIMED.  IN NO
+ * EVENT SHALL THE COPYRIGHT HOLDER(S) BE LIABLE FOR ANY DIRECT, INDIRECT,
+ * INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT
+ * LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, DATA, OR
+ * PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY THEORY OF
+ * LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT (INCLUDING NEGLIGENCE
+ * OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE OF THIS SOFTWARE, EVEN IF
+ * ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
+ *
+ * ***** 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.
+ *
+ * The Original Code is: adapted from jemalloc.
+ *
+ * ***** END GPL LICENSE BLOCK *****
+ */
+
+#ifndef __ATOMIC_OPS_UNIX_H__
+#define __ATOMIC_OPS_UNIX_H__
+
+#include "atomic_ops_utils.h"
+
+/******************************************************************************/
+/* 64-bit operations. */
+#if (LG_SIZEOF_PTR == 8 || LG_SIZEOF_INT == 8)
+#  if (defined(__GCC_HAVE_SYNC_COMPARE_AND_SWAP_8) || defined(JE_FORCE_SYNC_COMPARE_AND_SWAP_8))
+ATOMIC_INLINE uint64_t atomic_add_uint64(uint64_t *p, uint64_t x)
+{
+	return __sync_add_and_fetch(p, x);
+}
+
+ATOMIC_INLINE uint64_t atomic_sub_uint64(uint64_t *p, uint64_t x)
+{
+	return __sync_sub_and_fetch(p, x);
+}
+
+ATOMIC_INLINE uint64_t atomic_cas_uint64(uint64_t *v, uint64_t old, uint64_t _new)
+{
+	return __sync_val_compare_and_swap(v, old, _new);
+}
+#  elif (defined(__amd64__) || defined(__x86_64__))
+ATOMIC_INLINE uint64_t atomic_add_uint64(uint64_t *p, uint64_t x)
+{
+	asm volatile (
+	    "lock; xaddq %0, %1;"
+	    : "+r" (x), "=m" (*p) /* Outputs. */
+	    : "m" (*p) /* Inputs. */
+	    );
+	return x;
+}
+
+ATOMIC_INLINE uint64_t atomic_sub_uint64(uint64_t *p, uint64_t x)
+{
+	x = (uint64_t)(-(int64_t)x);
+	asm volatile (
+	    "lock; xaddq %0, %1;"
+	    : "+r" (x), "=m" (*p) /* Outputs. */
+	    : "m" (*p) /* Inputs. */
+	    );
+	return x;
+}
+
+ATOMIC_INLINE uint64_t atomic_cas_uint64(uint64_t *v, uint64_t old, uint64_t _new)
+{
+	uint64_t ret;
+	asm volatile (
+	    "lock; cmpxchgq %2,%1"
+	    : "=a" (ret), "+m" (*v)
+	    : "r" (_new), "0" (old)
+	    : "memory");
+	return ret;
+}
+#  else
+#    error "Missing implementation for 64-bit atomic operations"
+#  endif
+#endif
+
+/******************************************************************************/
+/* 32-bit operations. */
+#if (defined(__GCC_HAVE_SYNC_COMPARE_AND_SWAP_4) || defined(JE_FORCE_SYNC_COMPARE_AND_SWAP_4))
+ATOMIC_INLINE uint32_t atomic_add_uint32(uint32_t *p, uint32_t x)
+{
+	return __sync_add_and_fetch(p, x);
+}
+
+ATOMIC_INLINE uint32_t atomic_sub_uint32(uint32_t *p, uint32_t x)
+{
+	return __sync_sub_and_fetch(p, x);
+}
+
+ATOMIC_INLINE uint32_t atomic_cas_uint32(uint32_t *v, uint32_t old, uint32_t _new)
+{
+   return __sync_val_compare_and_swap(v, old, _new);
+}
+#elif (defined(__i386__) || defined(__amd64__) || defined(__x86_64__))
+ATOMIC_INLINE uint32_t atomic_add_uint32(uint32_t *p, uint32_t x)
+{
+	asm volatile (
+	    "lock; xaddl %0, %1;"
+	    : "+r" (x), "=m" (*p) /* Outputs. */
+	    : "m" (*p) /* Inputs. */
+	    );
+	return x;
+}
+
+ATOMIC_INLINE uint32_t atomic_sub_uint32(uint32_t *p, uint32_t x)
+{
+	x = (uint32_t)(-(int32_t)x);
+	asm volatile (
+	    "lock; xaddl %0, %1;"
+	    : "+r" (x), "=m" (*p) /* Outputs. */
+	    : "m" (*p) /* Inputs. */
+	    );
+	return x;
+}
+
+ATOMIC_INLINE uint32_t atomic_cas_uint32(uint32_t *v, uint32_t old, uint32_t _new)
+{
+	uint32_t ret;
+	asm volatile (
+	    "lock; cmpxchgl %2,%1"
+	    : "=a" (ret), "+m" (*v)
+	    : "r" (_new), "0" (old)
+	    : "memory");
+	return ret;
+}
+#else
+#  error "Missing implementation for 32-bit atomic operations"
+#endif
+
+#if (defined(__GCC_HAVE_SYNC_COMPARE_AND_SWAP_4) || defined(JE_FORCE_SYNC_COMPARE_AND_SWAP_4))
+ATOMIC_INLINE uint32_t atomic_fetch_and_add_uint32(uint32_t *p, uint32_t x)
+{
+	return __sync_fetch_and_add(p, x);
+}
+
+#else
+#  error "Missing implementation for 32-bit atomic operations"
+#endif
+
+/******************************************************************************/
+/* 8-bit operations. */
+#if (defined(__GCC_HAVE_SYNC_COMPARE_AND_SWAP_1) || defined(JE_FORCE_SYNC_COMPARE_AND_SWAP_1))
+ATOMIC_INLINE uint8_t atomic_fetch_and_and_uint8(uint8_t *p, uint8_t b)
+{
+	return __sync_fetch_and_and(p, b);
+}
+ATOMIC_INLINE uint8_t atomic_fetch_and_or_uint8(uint8_t *p, uint8_t b)
+{
+	return __sync_fetch_and_or(p, b);
+}
+#else
+#  error "Missing implementation for 8-bit atomic operations"
+#endif
+
+#endif /* __ATOMIC_OPS_UNIX_H__ */

intern/atomic/intern/atomic_ops_utils.h

@@ -0,0 +1,110 @@
+/*
+ * Original code from jemalloc with this license:
+ *
+ * Copyright (C) 2002-2013 Jason Evans <jasone@canonware.com>.
+ * All rights reserved.
+ * Copyright (C) 2007-2012 Mozilla Foundation.  All rights reserved.
+ * Copyright (C) 2009-2013 Facebook, Inc.  All rights reserved.
+ *
+ * Redistribution and use in source and binary forms, with or without
+ * modification, are permitted provided that the following conditions are met:
+ * 1. Redistributions of source code must retain the above copyright notice(s),
+ *    this list of conditions and the following disclaimer.
+ * 2. Redistributions in binary form must reproduce the above copyright notice(s),
+ *    this list of conditions and the following disclaimer in the documentation
+ *    and/or other materials provided with the distribution.
+ *
+ * THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDER(S) ``AS IS'' AND ANY EXPRESS
+ * OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE IMPLIED WARRANTIES OF
+ * MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE ARE DISCLAIMED.  IN NO
+ * EVENT SHALL THE COPYRIGHT HOLDER(S) BE LIABLE FOR ANY DIRECT, INDIRECT,
+ * INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT
+ * LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, DATA, OR
+ * PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY THEORY OF
+ * LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT (INCLUDING NEGLIGENCE
+ * OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE OF THIS SOFTWARE, EVEN IF
+ * ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
+ *
+ * ***** 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.
+ *
+ * The Original Code is: adapted from jemalloc.
+ *
+ * ***** END GPL LICENSE BLOCK *****
+ */
+
+#ifndef __ATOMIC_OPS_UTILS_H__
+#define __ATOMIC_OPS_UTILS_H__
+
+/* needed for int types */
+#include "../../../source/blender/blenlib/BLI_sys_types.h"
+#include <stdlib.h>
+#include <limits.h>
+
+#include <assert.h>
+
+/* little macro so inline keyword works */
+#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
+#endif
+
+#ifndef LIKELY
+#  ifdef __GNUC__
+#    define LIKELY(x)       __builtin_expect(!!(x), 1)
+#    define UNLIKELY(x)     __builtin_expect(!!(x), 0)
+#  else
+#    define LIKELY(x)       (x)
+#    define UNLIKELY(x)     (x)
+#  endif
+#endif
+
+#ifdef UINTPTR_MAX
+#  if (UINTPTR_MAX == 0xFFFFFFFF)
+#    define LG_SIZEOF_PTR 4
+#  elif (UINTPTR_MAX == 0xFFFFFFFFFFFFFFFF)
+#    define LG_SIZEOF_PTR 8
+#  endif
+#elif defined(__WORDSIZE)  /* Fallback for older glibc and cpp */
+#  if (__WORDSIZE == 32)
+#    define LG_SIZEOF_PTR 4
+#  elif (__WORDSIZE == 64)
+#    define LG_SIZEOF_PTR 8
+#  endif
+#endif
+
+#ifndef LG_SIZEOF_PTR
+#  error "Cannot find pointer size"
+#endif
+
+#if (UINT_MAX == 0xFFFFFFFF)
+#  define LG_SIZEOF_INT 4
+#elif (UINT_MAX == 0xFFFFFFFFFFFFFFFF)
+#  define LG_SIZEOF_INT 8
+#else
+#  error "Cannot find int size"
+#endif
+
+#endif /* __ATOMIC_OPS_UTILS_H__ */

intern/cycles/CMakeLists.txt

@@ -14,7 +14,15 @@ include(cmake/external_libs.cmake)
 # todo: this code could be refactored a bit to avoid duplication
 # note: CXX_HAS_SSE is needed in case passing SSE flags fails altogether (gcc-arm)
 
-if(NOT WITH_CPU_SSE)
+if(WITH_CYCLES_NATIVE_ONLY)
+	set(CXX_HAS_SSE FALSE)
+	set(CXX_HAS_AVX FALSE)
+	set(CXX_HAS_AVX2 FALSE)
+	add_definitions(
+		-DWITH_KERNEL_NATIVE
+	)
+	set(CMAKE_CXX_FLAGS "${CMAKE_CXX_FLAGS} -march=native")
+elseif(NOT WITH_CPU_SSE)
 	set(CXX_HAS_SSE FALSE)
 	set(CXX_HAS_AVX FALSE)
 	set(CXX_HAS_AVX2 FALSE)
@@ -145,7 +153,9 @@ set(WITH_CYCLES_DEVICE_MULTI TRUE)
 if(CYCLES_STANDALONE_REPOSITORY)
 	TEST_UNORDERED_MAP_SUPPORT()
 endif()
-if(HAVE_STD_UNORDERED_MAP_HEADER)
+if(WITH_CXX11)
+	add_definitions(-DCYCLES_STD_UNORDERED_MAP)
+elseif(HAVE_STD_UNORDERED_MAP_HEADER)
 	if(HAVE_UNORDERED_MAP_IN_STD_NAMESPACE)
 		add_definitions(-DCYCLES_STD_UNORDERED_MAP)
 	else()

intern/cycles/app/CMakeLists.txt

@@ -84,18 +84,12 @@ include_directories(SYSTEM ${INC_SYS})
 macro(cycles_target_link_libraries target)
 	target_link_libraries(${target} ${LIBRARIES})
 	if(WITH_CYCLES_OSL)
-		target_link_libraries_decoupled(${target} OSL_LIBRARIES)
-		if(MSVC)
-			target_link_libraries_debug(${target} "${LLVM_LIBRARIES_DEBUG}")
-			target_link_libraries_optimized(${target} "${LLVM_LIBRARIES}")
-		else()
-			target_link_libraries(${target} ${LLVM_LIBRARIES})
-		endif()
+		target_link_libraries(${target} ${OSL_LIBRARIES} ${LLVM_LIBRARIES})
 	endif()
-	target_link_libraries_decoupled(${target} OPENIMAGEIO_LIBRARIES)
-	target_link_libraries_decoupled(${target} OPENEXR_LIBRARIES)
 	target_link_libraries(
 		${target}
+		${OPENIMAGEIO_LIBRARIES}
+		${OPENEXR_LIBRARIES}
 		${PUGIXML_LIBRARIES}
 		${BOOST_LIBRARIES}
 		${CMAKE_DL_LIBS}

intern/cycles/app/cycles_standalone.cpp

@@ -32,6 +32,7 @@
 #include "util_string.h"
 #include "util_time.h"
 #include "util_transform.h"
+#include "util_version.h"
 
 #ifdef WITH_CYCLES_STANDALONE_GUI
 #include "util_view.h"
@@ -358,7 +359,7 @@ static void options_parse(int argc, const char **argv)
 
 	/* parse options */
 	ArgParse ap;
-	bool help = false, debug = false;
+	bool help = false, debug = false, version = false;
 	int verbosity = 1;
 
 	ap.options ("Usage: cycles [options] file.xml",
@@ -380,6 +381,7 @@ static void options_parse(int argc, const char **argv)
 		"--verbose %d", &verbosity, "Set verbosity of the logger",
 #endif
 		"--help", &help, "Print help message",
+		"--version", &version, "Print version number",
 		NULL);
 
 	if(ap.parse(argc, argv) < 0) {
@@ -406,6 +408,10 @@ static void options_parse(int argc, const char **argv)
 
 		exit(EXIT_SUCCESS);
 	}
+	else if(version) {
+		printf("%s\n", CYCLES_VERSION_STRING);
+		exit(EXIT_SUCCESS);
+	}
 	else if(help || options.filepath == "") {
 		ap.usage();
 		exit(EXIT_SUCCESS);

intern/cycles/app/cycles_xml.cpp

@@ -52,7 +52,7 @@ struct XMLReadState {
 	Scene *scene;		/* scene pointer */
 	Transform tfm;		/* current transform state */
 	bool smooth;		/* smooth normal state */
-	int shader;			/* current shader */
+	Shader *shader;		/* current shader */
 	string base;		/* base path to current file*/
 	float dicing_rate;	/* current dicing rate */
 	Mesh::DisplacementMethod displacement_method;
@@ -60,7 +60,7 @@ struct XMLReadState {
 	XMLReadState()
 	  : scene(NULL),
 	    smooth(false),
-	    shader(0),
+	    shader(NULL),
 	    dicing_rate(0.0f),
 	    displacement_method(Mesh::DISPLACE_BUMP)
 	{
@@ -230,6 +230,24 @@ static bool xml_read_enum(ustring *str, ShaderEnum& enm, pugi::xml_node node, co
 	return false;
 }
 
+static bool xml_read_enum_value(int *value, ShaderEnum& enm, pugi::xml_node node, const char *name)
+{
+	pugi::xml_attribute attr = node.attribute(name);
+
+	if(attr) {
+		ustring ustr(attr.value());
+
+		if(enm.exists(ustr)) {
+			*value = enm[ustr];
+			return true;
+		}
+		else
+			fprintf(stderr, "Unknown value \"%s\" for attribute \"%s\".\n", ustr.c_str(), name);
+	}
+
+	return false;
+}
+
 static ShaderSocketType xml_read_socket_type(pugi::xml_node node, const char *name)
 {
 	pugi::xml_attribute attr = node.attribute(name);
@@ -437,7 +455,7 @@ static void xml_read_shader_graph(const XMLReadState& state, Shader *shader, pug
 			/* Generate inputs/outputs from node sockets
 			 *
 			 * Note: ShaderInput/ShaderOutput store shallow string copies only!
-			 * Socket names must be stored in the extra lists instead. */
+			 * So we register them as ustring to ensure the pointer stays valid. */
 			/* read input values */
 			for(pugi::xml_node param = node.first_child(); param; param = param.next_sibling()) {
 				if(string_iequals(param.name(), "input")) {
@@ -449,8 +467,7 @@ static void xml_read_shader_graph(const XMLReadState& state, Shader *shader, pug
 					if(type == SHADER_SOCKET_UNDEFINED)
 						continue;
 					
-					osl->input_names.push_back(ustring(name));
-					osl->add_input(osl->input_names.back().c_str(), type);
+					osl->add_input(ustring(name).c_str(), type);
 				}
 				else if(string_iequals(param.name(), "output")) {
 					string name;
@@ -461,8 +478,7 @@ static void xml_read_shader_graph(const XMLReadState& state, Shader *shader, pug
 					if(type == SHADER_SOCKET_UNDEFINED)
 						continue;
 					
-					osl->output_names.push_back(ustring(name));
-					osl->add_output(osl->output_names.back().c_str(), type);
+					osl->add_output(ustring(name).c_str(), type);
 				}
 			}
 			
@@ -531,7 +547,24 @@ static void xml_read_shader_graph(const XMLReadState& state, Shader *shader, pug
 			snode = bump;
 		}
 		else if(string_iequals(node.name(), "mapping")) {
-			snode = new MappingNode();
+			MappingNode *map = new MappingNode();
+
+			TextureMapping *texmap = &map->tex_mapping;
+			xml_read_enum_value((int*) &texmap->type, TextureMapping::type_enum, node, "type");
+			xml_read_enum_value((int*) &texmap->projection, TextureMapping::projection_enum, node, "projection");
+			xml_read_enum_value((int*) &texmap->x_mapping, TextureMapping::mapping_enum, node, "x_mapping");
+			xml_read_enum_value((int*) &texmap->y_mapping, TextureMapping::mapping_enum, node, "y_mapping");
+			xml_read_enum_value((int*) &texmap->z_mapping, TextureMapping::mapping_enum, node, "z_mapping");
+			xml_read_bool(&texmap->use_minmax, node, "use_minmax");
+			if(texmap->use_minmax) {
+				xml_read_float3(&texmap->min, node, "min");
+				xml_read_float3(&texmap->max, node, "max");
+			}
+			xml_read_float3(&texmap->translation, node, "translation");
+			xml_read_float3(&texmap->rotation, node, "rotation");
+			xml_read_float3(&texmap->scale, node, "scale");
+
+			snode = map;
 		}
 		else if(string_iequals(node.name(), "anisotropic_bsdf")) {
 			AnisotropicBsdfNode *aniso = new AnisotropicBsdfNode();
@@ -867,7 +900,7 @@ static void xml_read_background(const XMLReadState& state, pugi::xml_node node)
 	xml_read_bool(&bg->transparent, node, "transparent");
 
 	/* Background Shader */
-	Shader *shader = state.scene->shaders[state.scene->default_background];
+	Shader *shader = state.scene->default_background;
 	
 	xml_read_bool(&shader->heterogeneous_volume, node, "heterogeneous_volume");
 	xml_read_int(&shader->volume_interpolation_method, node, "volume_interpolation_method");
@@ -906,7 +939,7 @@ static void xml_read_mesh(const XMLReadState& state, pugi::xml_node node)
 	mesh->used_shaders.push_back(state.shader);
 
 	/* read state */
-	int shader = state.shader;
+	int shader = 0;
 	bool smooth = state.smooth;
 
 	mesh->displacement_method = state.displacement_method;
@@ -1008,6 +1041,8 @@ static void xml_read_mesh(const XMLReadState& state, pugi::xml_node node)
 					fdata[2] = make_float3(UV[v2*2], UV[v2*2+1], 0.0);
 					fdata += 3;
 				}
+
+				index_offset += nverts[i];
 			}
 		}
 	}
@@ -1064,7 +1099,7 @@ static void xml_read_patch(const XMLReadState& state, pugi::xml_node node)
 		mesh->used_shaders.push_back(state.shader);
 
 		/* split */
-		SubdParams sdparams(mesh, state.shader, state.smooth);
+		SubdParams sdparams(mesh, 0, state.smooth);
 		xml_read_float(&sdparams.dicing_rate, node, "dicing_rate");
 
 		DiagSplit dsplit(sdparams);
@@ -1161,17 +1196,14 @@ static void xml_read_state(XMLReadState& state, pugi::xml_node node)
 	string shadername;
 
 	if(xml_read_string(&shadername, node, "shader")) {
-		int i = 0;
 		bool found = false;
 
 		foreach(Shader *shader, state.scene->shaders) {
 			if(shader->name == shadername) {
-				state.shader = i;
+				state.shader = shader;
 				found = true;
 				break;
 			}
-
-			i++;
 		}
 
 		if(!found)

intern/cycles/blender/addon/engine.py

@@ -50,10 +50,39 @@ def _workaround_buggy_drivers():
             _cycles.opencl_disable()
 
 
+def _parse_command_line():
+    import sys
+
+    argv = sys.argv
+    if "--" not in argv:
+        return
+
+    argv = argv[argv.index("--") + 1:]
+
+    num_resumable_chunks = None
+    current_resumable_chunk = None
+
+    # TODO(sergey): Add some nice error ptins if argument is not used properly.
+    idx = 0
+    while idx < len(argv) - 1:
+        arg = argv[idx]
+        if arg == '--cycles-resumable-num-chunks':
+            num_resumable_chunks = int(argv[idx + 1])
+        elif arg == '--cycles-resumable-current-chunk':
+            current_resumable_chunk = int(argv[idx + 1])
+        idx += 1
+
+    if num_resumable_chunks is not None and current_resumable_chunk is not None:
+        import _cycles
+        _cycles.set_resumable_chunks(num_resumable_chunks,
+                                     current_resumable_chunk)
+
+
 def init():
     import bpy
     import _cycles
     import os.path
+    import sys
 
     # Workaround possibly buggy legacy drivers which crashes on the OpenCL
     # device enumeration.
@@ -72,7 +101,7 @@ def init():
     user_path = os.path.dirname(os.path.abspath(bpy.utils.user_resource('CONFIG', '')))
 
     _cycles.init(path, user_path, bpy.app.background)
-
+    _parse_command_line()
 
 def exit():
     import _cycles

intern/cycles/blender/addon/properties.py

@@ -46,6 +46,12 @@ enum_displacement_methods = (
     ('BOTH', "Both", "Combination of displacement and bump mapping"),
     )
 
+enum_subdivision_types = (
+    ('NONE', "None", "No subdivision"),
+    ('LINEAR', "Linear", "Use linear subdivision"),
+    ('CATMULL_CLARK', "Catmull–Clark", "Use Catmull-Clark subdivision"),
+    )
+
 enum_bvh_types = (
     ('DYNAMIC_BVH', "Dynamic BVH", "Objects can be individually updated, at the cost of slower render time"),
     ('STATIC_BVH', "Static BVH", "Any object modification requires a complete BVH rebuild, but renders faster"),
@@ -187,57 +193,57 @@ class CyclesRenderSettings(bpy.types.PropertyGroup):
         cls.aa_samples = IntProperty(
                 name="AA Samples",
                 description="Number of antialiasing samples to render for each pixel",
-                min=1, max=10000,
+                min=1, max=2097151,
                 default=4,
                 )
         cls.preview_aa_samples = IntProperty(
                 name="AA Samples",
                 description="Number of antialiasing samples to render in the viewport, unlimited if 0",
-                min=0, max=10000,
+                min=0, max=2097151,
                 default=4,
                 )
         cls.diffuse_samples = IntProperty(
                 name="Diffuse Samples",
                 description="Number of diffuse bounce samples to render for each AA sample",
-                min=1, max=10000,
+                min=1, max=1024,
                 default=1,
                 )
         cls.glossy_samples = IntProperty(
                 name="Glossy Samples",
                 description="Number of glossy bounce samples to render for each AA sample",
-                min=1, max=10000,
+                min=1, max=1024,
                 default=1,
                 )
         cls.transmission_samples = IntProperty(
                 name="Transmission Samples",
                 description="Number of transmission bounce samples to render for each AA sample",
-                min=1, max=10000,
+                min=1, max=1024,
                 default=1,
                 )
         cls.ao_samples = IntProperty(
                 name="Ambient Occlusion Samples",
                 description="Number of ambient occlusion samples to render for each AA sample",
-                min=1, max=10000,
+                min=1, max=1024,
                 default=1,
                 )
         cls.mesh_light_samples = IntProperty(
                 name="Mesh Light Samples",
                 description="Number of mesh emission light samples to render for each AA sample",
-                min=1, max=10000,
+                min=1, max=1024,
                 default=1,
                 )
 
         cls.subsurface_samples = IntProperty(
                 name="Subsurface Samples",
                 description="Number of subsurface scattering samples to render for each AA sample",
-                min=1, max=10000,
+                min=1, max=1024,
                 default=1,
                 )
 
         cls.volume_samples = IntProperty(
                 name="Volume Samples",
                 description="Number of volume scattering samples to render for each AA sample",
-                min=1, max=10000,
+                min=1, max=1024,
                 default=1,
                 )
 
@@ -364,6 +370,26 @@ class CyclesRenderSettings(bpy.types.PropertyGroup):
                 min=2, max=65536
                 )
 
+        cls.dicing_rate = FloatProperty(
+                name="Dicing Rate",
+                description="Size of a micropolygon in pixels",
+                min=0.1, max=1000.0,
+                default=1.0,
+                )
+        cls.preview_dicing_rate = FloatProperty(
+                name="Preview Dicing Rate",
+                description="Size of a micropolygon in pixels during preview render",
+                min=0.1, max=1000.0,
+                default=8.0,
+                )
+
+        cls.max_subdivisions = IntProperty(
+                name="Max Subdivisions",
+                description="Stop subdividing when this level is reached even if the dice rate would produce finer tessellation",
+                min=0, max=16,
+                default=12,
+                )
+
         cls.film_exposure = FloatProperty(
                 name="Exposure",
                 description="Image brightness scale",
@@ -568,6 +594,8 @@ class CyclesRenderSettings(bpy.types.PropertyGroup):
         cls.debug_use_cpu_sse2 = BoolProperty(name="SSE2", default=True)
         cls.debug_use_qbvh = BoolProperty(name="QBVH", default=True)
 
+        cls.debug_use_cuda_adaptive_compile = BoolProperty(name="Adaptive Compile", default=False)
+
         cls.debug_opencl_kernel_type = EnumProperty(
             name="OpenCL Kernel Type",
             default='DEFAULT',
@@ -931,15 +959,16 @@ class CyclesMeshSettings(bpy.types.PropertyGroup):
                 items=enum_displacement_methods,
                 default='BUMP',
                 )
-        cls.use_subdivision = BoolProperty(
-                name="Use Subdivision",
-                description="Subdivide mesh for rendering",
-                default=False,
+        cls.subdivision_type = EnumProperty(
+                name="Subdivision Type",
+                description="Type of subdivision to use",
+                items=enum_subdivision_types,
+                default='NONE',
                 )
         cls.dicing_rate = FloatProperty(
                 name="Dicing Rate",
-                description="",
-                min=0.001, max=1000.0,
+                description="Multiplier for scene dicing rate",
+                min=0.1, max=1000.0,
                 default=1.0,
                 )
 

intern/cycles/blender/addon/ui.py

@@ -209,8 +209,8 @@ class CyclesRender_PT_sampling(CyclesButtonsPanel, Panel):
         draw_samples_info(layout, context)
 
 
-class CyclesRender_PT_volume_sampling(CyclesButtonsPanel, Panel):
-    bl_label = "Volume Sampling"
+class CyclesRender_PT_geometery(CyclesButtonsPanel, Panel):
+    bl_label = "Geometry"
     bl_options = {'DEFAULT_CLOSED'}
 
     def draw(self, context):
@@ -219,11 +219,30 @@ class CyclesRender_PT_volume_sampling(CyclesButtonsPanel, Panel):
         scene = context.scene
         cscene = scene.cycles
 
-        row = layout.row()
-        row.label("Heterogeneous:")
-        row = layout.row()
-        row.prop(cscene, "volume_step_size")
-        row.prop(cscene, "volume_max_steps")
+        if cscene.feature_set == 'EXPERIMENTAL':
+            split = layout.split()
+
+            col = split.column()
+
+            sub = col.column(align=True)
+            sub.label("Volume Sampling:")
+            sub.prop(cscene, "volume_step_size")
+            sub.prop(cscene, "volume_max_steps")
+
+            col = split.column()
+
+            sub = col.column(align=True)
+            sub.label("Subdivision Rate:")
+            sub.prop(cscene, "dicing_rate", text="Render")
+            sub.prop(cscene, "preview_dicing_rate", text="Preview")
+            sub.separator()
+            sub.prop(cscene, "max_subdivisions")
+        else:
+            row = layout.row()
+            row.label("Volume Sampling:")
+            row = layout.row()
+            row.prop(cscene, "volume_step_size")
+            row.prop(cscene, "volume_max_steps")
 
 
 class CyclesRender_PT_light_paths(CyclesButtonsPanel, Panel):
@@ -681,10 +700,20 @@ class Cycles_PT_mesh_displacement(CyclesButtonsPanel, Panel):
         elif mball:
             cdata = mball.cycles
 
-        layout.prop(cdata, "displacement_method", text="Method")
-        layout.prop(cdata, "use_subdivision")
-        layout.prop(cdata, "dicing_rate")
+        split = layout.split()
 
+        col = split.column()
+        sub = col.column(align=True)
+        sub.label(text="Displacment:")
+        sub.prop(cdata, "displacement_method", text="")
+
+        col = split.column()
+        sub = col.column(align=True)
+        sub.label(text="Subdivision:")
+        sub.prop(cdata, "subdivision_type", text="")
+
+        if cdata.subdivision_type != 'NONE':
+            sub.prop(cdata, "dicing_rate")
 
 class CyclesObject_PT_motion_blur(CyclesButtonsPanel, Panel):
     bl_label = "Motion Blur"
@@ -1505,7 +1534,7 @@ class CyclesRender_PT_debug(CyclesButtonsPanel, Panel):
 
     @classmethod
     def poll(cls, context):
-        return bpy.app.debug_value == 256
+        return CyclesButtonsPanel.poll(context) and bpy.app.debug_value == 256
 
     def draw(self, context):
         layout = self.layout
@@ -1524,6 +1553,10 @@ class CyclesRender_PT_debug(CyclesButtonsPanel, Panel):
         row.prop(cscene, "debug_use_cpu_avx2", toggle=True)
         col.prop(cscene, "debug_use_qbvh")
 
+        col = layout.column()
+        col.label('CUDA Flags:')
+        col.prop(cscene, "debug_use_cuda_adaptive_compile")
+
         col = layout.column()
         col.label('OpenCL Flags:')
         col.prop(cscene, "debug_opencl_kernel_type", text="Kernel")

intern/cycles/blender/blender_camera.cpp

@@ -65,6 +65,9 @@ struct BlenderCamera {
 	bool use_spherical_stereo;
 	float interocular_distance;
 	float convergence_distance;
+	bool use_pole_merge;
+	float pole_merge_angle_from;
+	float pole_merge_angle_to;
 
 	enum { AUTO, HORIZONTAL, VERTICAL } sensor_fit;
 	float sensor_width;
@@ -183,6 +186,10 @@ static void blender_camera_from_object(BlenderCamera *bcam,
 		}
 		bcam->use_spherical_stereo = b_engine.use_spherical_stereo(b_ob);
 
+		bcam->use_pole_merge = b_camera.stereo().use_pole_merge();
+		bcam->pole_merge_angle_from = b_camera.stereo().pole_merge_angle_from();
+		bcam->pole_merge_angle_to = b_camera.stereo().pole_merge_angle_to();
+
 		bcam->ortho_scale = b_camera.ortho_scale();
 
 		bcam->lens = b_camera.lens();
@@ -360,6 +367,12 @@ static void blender_camera_sync(Camera *cam, BlenderCamera *bcam, int width, int
 	blender_camera_viewplane(bcam, width, height,
 		&cam->viewplane, &aspectratio, &sensor_size);
 
+	cam->width = bcam->full_width;
+	cam->height = bcam->full_height;
+
+	cam->full_width = width;
+	cam->full_height = height;
+
 	/* panorama sensor */
 	if(bcam->type == CAMERA_PANORAMA && bcam->panorama_type == PANORAMA_FISHEYE_EQUISOLID) {
 		float fit_xratio = (float)bcam->full_width*bcam->pixelaspect.x;
@@ -421,6 +434,10 @@ static void blender_camera_sync(Camera *cam, BlenderCamera *bcam, int width, int
 			cam->stereo_eye = Camera::STEREO_NONE;
 	}
 
+	cam->use_pole_merge = bcam->use_pole_merge;
+	cam->pole_merge_angle_from = bcam->pole_merge_angle_from;
+	cam->pole_merge_angle_to = bcam->pole_merge_angle_to;
+
 	/* anamorphic lens bokeh */
 	cam->aperture_ratio = bcam->aperture_ratio;
 

intern/cycles/blender/blender_curves.cpp

@@ -138,8 +138,7 @@ bool ObtainCacheParticleData(Mesh *mesh, BL::Mesh *b_mesh, BL::Object *b_ob, Par
 			BL::ParticleSettings b_part((const PointerRNA)b_psys.settings().ptr);
 
 			if((b_part.render_type() == BL::ParticleSettings::render_type_PATH) && (b_part.type() == BL::ParticleSettings::type_HAIR)) {
-				int mi = clamp(b_part.material()-1, 0, mesh->used_shaders.size()-1);
-				int shader = mesh->used_shaders[mi];
+				int shader = clamp(b_part.material()-1, 0, mesh->used_shaders.size()-1);
 				int draw_step = background ? b_part.render_step() : b_part.draw_step();
 				int totparts = b_psys.particles.length();
 				int totchild = background ? b_psys.child_particles.length() : (int)((float)b_psys.child_particles.length() * (float)b_part.draw_percentage() / 100.0f);