Splines: Fix curve to mesh node issues

- The expected flat axis of profile objects is now the Z axis, like
    the existing curve objects.
 - Simplify/speed up the "sharp edge" marking logic.

.clang-format

@@ -255,7 +255,6 @@ ForEachMacros:
   - SCULPT_VERTEX_DUPLICATES_AND_NEIGHBORS_ITER_BEGIN
   - SCULPT_VERTEX_NEIGHBORS_ITER_BEGIN
   - SEQ_ALL_BEGIN
-  - SEQ_ITERATOR_FOREACH
   - SURFACE_QUAD_ITER_BEGIN
   - foreach
   - ED_screen_areas_iter

build_files/build_environment/cmake/blosc.cmake

@@ -29,7 +29,7 @@ set(BLOSC_EXTRA_ARGS
   -DCMAKE_POSITION_INDEPENDENT_CODE=ON
 )
 
-# Prevent blosc from including its own local copy of zlib in the object file
+# Prevent blosc from including it's own local copy of zlib in the object file
 # and cause linker errors with everybody else.
 set(BLOSC_EXTRA_ARGS ${BLOSC_EXTRA_ARGS}
   -DPREFER_EXTERNAL_ZLIB=ON

build_files/build_environment/cmake/versions.cmake

@@ -43,7 +43,7 @@ set(JPEG_FILE libjpeg-turbo-${JPEG_VERSION}.tar.gz)
 set(BOOST_VERSION 1.73.0)
 set(BOOST_VERSION_NODOTS 1_73_0)
 set(BOOST_VERSION_NODOTS_SHORT 1_73)
-set(BOOST_URI https://boostorg.jfrog.io/artifactory/main/release/${BOOST_VERSION}/source/boost_${BOOST_VERSION_NODOTS}.tar.gz)
+set(BOOST_URI https://dl.bintray.com/boostorg/release/${BOOST_VERSION}/source/boost_${BOOST_VERSION_NODOTS}.tar.gz)
 set(BOOST_HASH 4036cd27ef7548b8d29c30ea10956196)
 set(BOOST_HASH_TYPE MD5)
 set(BOOST_FILE boost_${BOOST_VERSION_NODOTS}.tar.gz)
@@ -297,10 +297,10 @@ set(OPENJPEG_HASH 63f5a4713ecafc86de51bfad89cc07bb788e9bba24ebbf0c4ca637621aadb6
 set(OPENJPEG_HASH_TYPE SHA256)
 set(OPENJPEG_FILE openjpeg-v${OPENJPEG_VERSION}.tar.gz)
 
-set(FFMPEG_VERSION 4.4)
+set(FFMPEG_VERSION 4.2.3)
 set(FFMPEG_URI http://ffmpeg.org/releases/ffmpeg-${FFMPEG_VERSION}.tar.bz2)
-set(FFMPEG_HASH 42093549751b582cf0f338a21a3664f52e0a9fbe0d238d3c992005e493607d0e)
-set(FFMPEG_HASH_TYPE SHA256)
+set(FFMPEG_HASH 695fad11f3baf27784e24cb0e977b65a)
+set(FFMPEG_HASH_TYPE MD5)
 set(FFMPEG_FILE ffmpeg-${FFMPEG_VERSION}.tar.bz2)
 
 set(FFTW_VERSION 3.3.8)

build_files/build_environment/install_deps.sh

@@ -37,7 +37,7 @@ if [ $USE_DEBUG_TRAP -ne 0 ]; then
   trap 'err_report $LINENO' ERR
 fi
 
-# Noisy, show every line that runs with its line number.
+# Noisy, show every line that runs with it's line number.
 if [ $USE_DEBUG_LOG -ne 0 ]; then
   PS4='\e[0;33m$(printf %4d ${LINENO}):\e\033[0m '
   set -x
@@ -563,9 +563,9 @@ OIDN_SKIP=false
 
 ISPC_VERSION="1.14.1"
 
-FFMPEG_VERSION="4.4"
-FFMPEG_VERSION_SHORT="4.4"
-FFMPEG_VERSION_MIN="4.4"
+FFMPEG_VERSION="4.2.3"
+FFMPEG_VERSION_SHORT="4.2"
+FFMPEG_VERSION_MIN="3.0"
 FFMPEG_VERSION_MAX="5.0"
 FFMPEG_FORCE_BUILD=false
 FFMPEG_FORCE_REBUILD=false

build_files/build_environment/patches/cmake/modules/FindIlmBase.cmake

@@ -20,7 +20,7 @@
 # ILMBASE_LIBRARIES   - list of libraries to link against when using IlmBase.
 # ILMBASE_FOUND       - True if IlmBase was found.
 
-# Other standard issue macros
+# Other standarnd issue macros
 include(FindPackageHandleStandardArgs)
 include(FindPackageMessage)
 include(SelectLibraryConfigurations)

build_files/build_environment/patches/cmake/modules/FindOpenEXR.cmake

@@ -22,7 +22,7 @@
 #                       These are defined by the FindIlmBase module.
 # OPENEXR_FOUND       - True if OpenEXR was found.
 
-# Other standard issue macros
+# Other standarnd issue macros
 include(SelectLibraryConfigurations)
 include(FindPackageHandleStandardArgs)
 include(FindPackageMessage)

build_files/cmake/buildinfo.cmake

@@ -79,7 +79,7 @@ if(EXISTS ${SOURCE_DIR}/.git)
                     ERROR_QUIET)
     if(NOT _git_below_check STREQUAL "")
       # If there're commits between HEAD and upstream this means
-      # that we're reset-ed to older revision. Use its hash then.
+      # that we're reset-ed to older revision. Use it's hash then.
       execute_process(COMMAND git rev-parse --short=12 HEAD
                       WORKING_DIRECTORY ${SOURCE_DIR}
                       OUTPUT_VARIABLE MY_WC_HASH

build_files/cmake/config/blender_release.cmake

@@ -56,6 +56,10 @@ set(WITH_TBB                 ON  CACHE BOOL "" FORCE)
 set(WITH_USD                 ON  CACHE BOOL "" FORCE)
 
 set(WITH_MEM_JEMALLOC          ON  CACHE BOOL "" FORCE)
+set(WITH_CYCLES_CUDA_BINARIES  ON  CACHE BOOL "" FORCE)
+set(WITH_CYCLES_CUBIN_COMPILER OFF CACHE BOOL "" FORCE)
+set(CYCLES_CUDA_BINARIES_ARCH sm_30;sm_35;sm_37;sm_50;sm_52;sm_60;sm_61;sm_70;sm_75;sm_86;compute_75 CACHE STRING "" FORCE)
+set(WITH_CYCLES_DEVICE_OPTIX   ON CACHE BOOL "" FORCE)
 
 # platform dependent options
 if(APPLE)
@@ -76,8 +80,4 @@ if(UNIX AND NOT APPLE)
 endif()
 if(NOT APPLE)
   set(WITH_XR_OPENXR              ON  CACHE BOOL "" FORCE)
-
-  set(WITH_CYCLES_DEVICE_OPTIX    ON  CACHE BOOL "" FORCE)
-  set(WITH_CYCLES_CUDA_BINARIES   ON  CACHE BOOL "" FORCE)
-  set(WITH_CYCLES_CUBIN_COMPILER  OFF CACHE BOOL "" FORCE)
 endif()

build_files/cmake/packaging.cmake

@@ -104,8 +104,8 @@ if(WIN32)
   set(CPACK_WIX_LIGHT_EXTRA_FLAGS -dcl:medium)
 endif()
 
-set(CPACK_PACKAGE_EXECUTABLES "blender-launcher" "blender")
-set(CPACK_CREATE_DESKTOP_LINKS "blender-launcher" "blender")
+set(CPACK_PACKAGE_EXECUTABLES "blender" "blender")
+set(CPACK_CREATE_DESKTOP_LINKS "blender" "blender")
 
 include(CPack)
 

build_files/cmake/platform/platform_apple.cmake

@@ -20,6 +20,12 @@
 
 # Libraries configuration for Apple.
 
+if("${CMAKE_OSX_ARCHITECTURES}" STREQUAL "arm64")
+  set(MACOSX_DEPLOYMENT_TARGET 11.00)
+else()
+  set(MACOSX_DEPLOYMENT_TARGET 10.13)
+endif()
+
 macro(find_package_wrapper)
 # do nothing, just satisfy the macro
 endmacro()

build_files/cmake/platform/platform_apple_xcode.cmake

@@ -168,15 +168,21 @@ endif()
 unset(OSX_SDKROOT)
 
 
+# 10.13 is our min. target, if you use higher sdk, weak linking happens
 if("${CMAKE_OSX_ARCHITECTURES}" STREQUAL "arm64")
-  # M1 chips run Big Sur onwards.
   set(OSX_MIN_DEPLOYMENT_TARGET 11.00)
 else()
-  # 10.13 is our min. target, if you use higher sdk, weak linking happens
   set(OSX_MIN_DEPLOYMENT_TARGET 10.13)
 endif()
 
-set(CMAKE_OSX_DEPLOYMENT_TARGET "${OSX_MIN_DEPLOYMENT_TARGET}" CACHE STRING "" FORCE)
+if(CMAKE_OSX_DEPLOYMENT_TARGET)
+  if(${CMAKE_OSX_DEPLOYMENT_TARGET} VERSION_LESS ${OSX_MIN_DEPLOYMENT_TARGET})
+    message(STATUS "Setting deployment target to ${OSX_MIN_DEPLOYMENT_TARGET}, lower versions are not supported")
+    set(CMAKE_OSX_DEPLOYMENT_TARGET "${OSX_MIN_DEPLOYMENT_TARGET}" CACHE STRING "" FORCE)
+  endif()
+else()
+  set(CMAKE_OSX_DEPLOYMENT_TARGET "${OSX_MIN_DEPLOYMENT_TARGET}" CACHE STRING "" FORCE)
+endif()
 
 if(NOT ${CMAKE_GENERATOR} MATCHES "Xcode")
   # Force CMAKE_OSX_DEPLOYMENT_TARGET for makefiles, will not work else (CMake bug?)

build_files/cmake/platform/platform_win32.cmake

@@ -119,7 +119,7 @@ string(APPEND CMAKE_MODULE_LINKER_FLAGS " /SAFESEH:NO /ignore:4099")
 list(APPEND PLATFORM_LINKLIBS
   ws2_32 vfw32 winmm kernel32 user32 gdi32 comdlg32 Comctl32 version
   advapi32 shfolder shell32 ole32 oleaut32 uuid psapi Dbghelp Shlwapi
-  pathcch Shcore
+  pathcch
 )
 
 if(WITH_INPUT_IME)
@@ -144,8 +144,8 @@ add_definitions(-D_ALLOW_KEYWORD_MACROS)
 # that both /GR and /GR- are specified.
 remove_cc_flag("/GR")
 
-# Make the Windows 8.1 API available for use.
-add_definitions(-D_WIN32_WINNT=0x603)
+# We want to support Windows 7 level ABI
+add_definitions(-D_WIN32_WINNT=0x601)
 include(build_files/cmake/platform/platform_win32_bundle_crt.cmake)
 remove_cc_flag("/MDd" "/MD" "/Zi")
 

doc/blender_file_format/BlendFileReader.py

@@ -85,7 +85,7 @@ def openBlendFile(filename):
     '''
     handle = open(filename, 'rb')
     magic = ReadString(handle, 7)
-    if magic in {"BLENDER", "BULLETf"}:
+    if magic in ("BLENDER", "BULLETf"):
         log.debug("normal blendfile detected")
         handle.seek(0, os.SEEK_SET)
         return handle
@@ -137,7 +137,7 @@ class BlendFile:
         fileblock = BlendFileBlock(handle, self)
         found_dna_block = False
         while not found_dna_block:
-            if fileblock.Header.Code in {"DNA1", "SDNA"}:
+            if fileblock.Header.Code in ("DNA1", "SDNA"):
                 self.Catalog = DNACatalog(self.Header, handle)
                 found_dna_block = True
             else:

doc/python_api/examples/bpy.types.RenderEngine.py

@@ -4,9 +4,7 @@ Simple Render Engine
 """
 
 import bpy
-import array
-import gpu
-from gpu_extras.presets import draw_texture_2d
+import bgl
 
 
 class CustomRenderEngine(bpy.types.RenderEngine):
@@ -102,7 +100,8 @@ class CustomRenderEngine(bpy.types.RenderEngine):
         dimensions = region.width, region.height
 
         # Bind shader that converts from scene linear to display space,
-        gpu.state.blend_set('ALPHA_PREMULT')
+        bgl.glEnable(bgl.GL_BLEND)
+        bgl.glBlendFunc(bgl.GL_ONE, bgl.GL_ONE_MINUS_SRC_ALPHA)
         self.bind_display_space_shader(scene)
 
         if not self.draw_data or self.draw_data.dimensions != dimensions:
@@ -111,7 +110,7 @@ class CustomRenderEngine(bpy.types.RenderEngine):
         self.draw_data.draw()
 
         self.unbind_display_space_shader()
-        gpu.state.blend_set('NONE')
+        bgl.glDisable(bgl.GL_BLEND)
 
 
 class CustomDrawData:
@@ -120,21 +119,68 @@ class CustomDrawData:
         self.dimensions = dimensions
         width, height = dimensions
 
-        pixels = width * height * array.array('f', [0.1, 0.2, 0.1, 1.0])
-        pixels = gpu.types.Buffer('FLOAT', width * height * 4, pixels)
+        pixels = [0.1, 0.2, 0.1, 1.0] * width * height
+        pixels = bgl.Buffer(bgl.GL_FLOAT, width * height * 4, pixels)
 
         # Generate texture
-        self.texture = gpu.types.GPUTexture((width, height), format='RGBA16F', data=pixels)
+        self.texture = bgl.Buffer(bgl.GL_INT, 1)
+        bgl.glGenTextures(1, self.texture)
+        bgl.glActiveTexture(bgl.GL_TEXTURE0)
+        bgl.glBindTexture(bgl.GL_TEXTURE_2D, self.texture[0])
+        bgl.glTexImage2D(bgl.GL_TEXTURE_2D, 0, bgl.GL_RGBA16F, width, height, 0, bgl.GL_RGBA, bgl.GL_FLOAT, pixels)
+        bgl.glTexParameteri(bgl.GL_TEXTURE_2D, bgl.GL_TEXTURE_MIN_FILTER, bgl.GL_LINEAR)
+        bgl.glTexParameteri(bgl.GL_TEXTURE_2D, bgl.GL_TEXTURE_MAG_FILTER, bgl.GL_LINEAR)
+        bgl.glBindTexture(bgl.GL_TEXTURE_2D, 0)
 
-        # Note: This is just a didactic example.
-        # In this case it would be more convenient to fill the texture with:
-        # self.texture.clear('FLOAT', value=[0.1, 0.2, 0.1, 1.0])
+        # Bind shader that converts from scene linear to display space,
+        # use the scene's color management settings.
+        shader_program = bgl.Buffer(bgl.GL_INT, 1)
+        bgl.glGetIntegerv(bgl.GL_CURRENT_PROGRAM, shader_program)
+
+        # Generate vertex array
+        self.vertex_array = bgl.Buffer(bgl.GL_INT, 1)
+        bgl.glGenVertexArrays(1, self.vertex_array)
+        bgl.glBindVertexArray(self.vertex_array[0])
+
+        texturecoord_location = bgl.glGetAttribLocation(shader_program[0], "texCoord")
+        position_location = bgl.glGetAttribLocation(shader_program[0], "pos")
+
+        bgl.glEnableVertexAttribArray(texturecoord_location)
+        bgl.glEnableVertexAttribArray(position_location)
+
+        # Generate geometry buffers for drawing textured quad
+        position = [0.0, 0.0, width, 0.0, width, height, 0.0, height]
+        position = bgl.Buffer(bgl.GL_FLOAT, len(position), position)
+        texcoord = [0.0, 0.0, 1.0, 0.0, 1.0, 1.0, 0.0, 1.0]
+        texcoord = bgl.Buffer(bgl.GL_FLOAT, len(texcoord), texcoord)
+
+        self.vertex_buffer = bgl.Buffer(bgl.GL_INT, 2)
+
+        bgl.glGenBuffers(2, self.vertex_buffer)
+        bgl.glBindBuffer(bgl.GL_ARRAY_BUFFER, self.vertex_buffer[0])
+        bgl.glBufferData(bgl.GL_ARRAY_BUFFER, 32, position, bgl.GL_STATIC_DRAW)
+        bgl.glVertexAttribPointer(position_location, 2, bgl.GL_FLOAT, bgl.GL_FALSE, 0, None)
+
+        bgl.glBindBuffer(bgl.GL_ARRAY_BUFFER, self.vertex_buffer[1])
+        bgl.glBufferData(bgl.GL_ARRAY_BUFFER, 32, texcoord, bgl.GL_STATIC_DRAW)
+        bgl.glVertexAttribPointer(texturecoord_location, 2, bgl.GL_FLOAT, bgl.GL_FALSE, 0, None)
+
+        bgl.glBindBuffer(bgl.GL_ARRAY_BUFFER, 0)
+        bgl.glBindVertexArray(0)
 
     def __del__(self):
-        del self.texture
+        bgl.glDeleteBuffers(2, self.vertex_buffer)
+        bgl.glDeleteVertexArrays(1, self.vertex_array)
+        bgl.glBindTexture(bgl.GL_TEXTURE_2D, 0)
+        bgl.glDeleteTextures(1, self.texture)
 
     def draw(self):
-        draw_texture_2d(self.texture, (0, 0), self.texture.width, self.texture.height)
+        bgl.glActiveTexture(bgl.GL_TEXTURE0)
+        bgl.glBindTexture(bgl.GL_TEXTURE_2D, self.texture[0])
+        bgl.glBindVertexArray(self.vertex_array[0])
+        bgl.glDrawArrays(bgl.GL_TRIANGLE_FAN, 0, 4)
+        bgl.glBindVertexArray(0)
+        bgl.glBindTexture(bgl.GL_TEXTURE_2D, 0)
 
 
 # RenderEngines also need to tell UI Panels that they are compatible with.

doc/python_api/examples/gpu.4.py

@@ -4,6 +4,7 @@ Mesh with Random Vertex Colors
 """
 import bpy
 import gpu
+import bgl
 import numpy as np
 from random import random
 from gpu_extras.batch import batch_for_shader
@@ -30,10 +31,9 @@ batch = batch_for_shader(
 
 
 def draw():
-    gpu.state.depth_test_set('LESS_EQUAL')
-    gpu.state.depth_mask_set(True)
+    bgl.glEnable(bgl.GL_DEPTH_TEST)
     batch.draw(shader)
-    gpu.state.depth_mask_set(False)
+    bgl.glDisable(bgl.GL_DEPTH_TEST)
 
 
 bpy.types.SpaceView3D.draw_handler_add(draw, (), 'WINDOW', 'POST_VIEW')

doc/python_api/examples/gpu.6.py

@@ -6,11 +6,11 @@ To use this example you have to provide an image that should be displayed.
 """
 import bpy
 import gpu
+import bgl
 from gpu_extras.batch import batch_for_shader
 
 IMAGE_NAME = "Untitled"
 image = bpy.data.images[IMAGE_NAME]
-texture = gpu.texture.from_image(image)
 
 shader = gpu.shader.from_builtin('2D_IMAGE')
 batch = batch_for_shader(
@@ -21,9 +21,16 @@ batch = batch_for_shader(
     },
 )
 
+if image.gl_load():
+    raise Exception()
+
+
 def draw():
+    bgl.glActiveTexture(bgl.GL_TEXTURE0)
+    bgl.glBindTexture(bgl.GL_TEXTURE_2D, image.bindcode)
+
     shader.bind()
-    shader.uniform_sampler("image", texture)
+    shader.uniform_int("image", 0)
     batch.draw(shader)
 
 

doc/python_api/examples/gpu.7.py

@@ -9,6 +9,7 @@ Generate a texture using Offscreen Rendering
 """
 import bpy
 import gpu
+import bgl
 from mathutils import Matrix
 from gpu_extras.batch import batch_for_shader
 from gpu_extras.presets import draw_circle_2d
@@ -19,8 +20,8 @@ from gpu_extras.presets import draw_circle_2d
 offscreen = gpu.types.GPUOffScreen(512, 512)
 
 with offscreen.bind():
-    fb = gpu.state.active_framebuffer_get()
-    fb.clear(color=(0.0, 0.0, 0.0, 0.0))
+    bgl.glClearColor(0.0, 0.0, 0.0, 0.0)
+    bgl.glClear(bgl.GL_COLOR_BUFFER_BIT)
     with gpu.matrix.push_pop():
         # reset matrices -> use normalized device coordinates [-1, 1]
         gpu.matrix.load_matrix(Matrix.Identity(4))
@@ -74,10 +75,13 @@ batch = batch_for_shader(
 
 
 def draw():
+    bgl.glActiveTexture(bgl.GL_TEXTURE0)
+    bgl.glBindTexture(bgl.GL_TEXTURE_2D, offscreen.color_texture)
+
     shader.bind()
     shader.uniform_float("modelMatrix", Matrix.Translation((1, 2, 3)) @ Matrix.Scale(3, 4))
     shader.uniform_float("viewProjectionMatrix", bpy.context.region_data.perspective_matrix)
-    shader.uniform_sampler("image", offscreen.texture_color)
+    shader.uniform_float("image", 0)
     batch.draw(shader)
 
 

doc/python_api/examples/gpu.8.py

@@ -7,10 +7,11 @@ If it already exists, it will override the existing one.
 
 Currently almost all of the execution time is spent in the last line.
 In the future this will hopefully be solved by implementing the Python buffer protocol
-for :class:`gpu.types.Buffer` and :class:`bpy.types.Image.pixels` (aka ``bpy_prop_array``).
+for :class:`bgl.Buffer` and :class:`bpy.types.Image.pixels` (aka ``bpy_prop_array``).
 """
 import bpy
 import gpu
+import bgl
 import random
 from mathutils import Matrix
 from gpu_extras.presets import draw_circle_2d
@@ -24,8 +25,8 @@ RING_AMOUNT = 10
 offscreen = gpu.types.GPUOffScreen(WIDTH, HEIGHT)
 
 with offscreen.bind():
-    fb = gpu.state.active_framebuffer_get()
-    fb.clear(color=(0.0, 0.0, 0.0, 0.0))
+    bgl.glClearColor(0.0, 0.0, 0.0, 0.0)
+    bgl.glClear(bgl.GL_COLOR_BUFFER_BIT)
     with gpu.matrix.push_pop():
         # reset matrices -> use normalized device coordinates [-1, 1]
         gpu.matrix.load_matrix(Matrix.Identity(4))
@@ -36,7 +37,9 @@ with offscreen.bind():
                 (random.uniform(-1, 1), random.uniform(-1, 1)),
                 (1, 1, 1, 1), random.uniform(0.1, 1), 20)
 
-    buffer = fb.read_color(0, 0, WIDTH, HEIGHT, 4, 0, 'UBYTE')
+    buffer = bgl.Buffer(bgl.GL_BYTE, WIDTH * HEIGHT * 4)
+    bgl.glReadBuffer(bgl.GL_BACK)
+    bgl.glReadPixels(0, 0, WIDTH, HEIGHT, bgl.GL_RGBA, bgl.GL_UNSIGNED_BYTE, buffer)
 
 offscreen.free()
 
@@ -45,6 +48,4 @@ if not IMAGE_NAME in bpy.data.images:
     bpy.data.images.new(IMAGE_NAME, WIDTH, HEIGHT)
 image = bpy.data.images[IMAGE_NAME]
 image.scale(WIDTH, HEIGHT)
-
-buffer.dimensions = WIDTH * HEIGHT * 4
 image.pixels = [v / 255 for v in buffer]

doc/python_api/examples/gpu.9.py

@@ -7,6 +7,7 @@ You could also make this independent of a specific camera,
 but Blender does not expose good functions to create view and projection matrices yet.
 """
 import bpy
+import bgl
 import gpu
 from gpu_extras.presets import draw_texture_2d
 
@@ -33,8 +34,8 @@ def draw():
         view_matrix,
         projection_matrix)
 
-    gpu.state.depth_mask_set(False)
-    draw_texture_2d(offscreen.texture_color, (10, 10), WIDTH, HEIGHT)
+    bgl.glDisable(bgl.GL_DEPTH_TEST)
+    draw_texture_2d(offscreen.color_texture, (10, 10), WIDTH, HEIGHT)
 
 
 bpy.types.SpaceView3D.draw_handler_add(draw, (), 'WINDOW', 'POST_PIXEL')

doc/python_api/examples/mathutils.Matrix.LocRotScale.py

@@ -1,5 +0,0 @@
-# Compute local object transformation matrix:
-if obj.rotation_mode == 'QUATERNION':
-    matrix = mathutils.Matrix.LocRotScale(obj.location, obj.rotation_quaternion, obj.scale)
-else:
-    matrix = mathutils.Matrix.LocRotScale(obj.location, obj.rotation_euler, obj.scale)

doc/python_api/examples/mathutils.Matrix.py

@@ -14,14 +14,10 @@ mat_rot = mathutils.Matrix.Rotation(math.radians(45.0), 4, 'X')
 mat_out = mat_loc @ mat_rot @ mat_sca
 print(mat_out)
 
-# extract components back out of the matrix as two vectors and a quaternion
+# extract components back out of the matrix
 loc, rot, sca = mat_out.decompose()
 print(loc, rot, sca)
 
-# recombine extracted components
-mat_out2 = mathutils.Matrix.LocRotScale(loc, rot, sca)
-print(mat_out2)
-
 # it can also be useful to access components of a matrix directly
 mat = mathutils.Matrix()
 mat[0][0], mat[1][0], mat[2][0] = 0.0, 1.0, 2.0

doc/python_api/sphinx_doc_gen.py

@@ -545,13 +545,6 @@ def range_str(val):
 
 
 def example_extract_docstring(filepath):
-    '''
-    Return (text, line_no, line_no_has_content) where:
-    - ``text`` is the doc-string text.
-    - ``line_no`` is the line the doc-string text ends.
-    - ``line_no_has_content`` when False, this file only contains a doc-string.
-      There is no need to include the remainder.
-    '''
     file = open(filepath, "r", encoding="utf-8")
     line = file.readline()
     line_no = 0
@@ -560,7 +553,7 @@ def example_extract_docstring(filepath):
         line_no += 1
     else:
         file.close()
-        return "", 0, True
+        return "", 0, False
 
     for line in file:
         line_no += 1
@@ -954,7 +947,7 @@ def pymodule2sphinx(basepath, module_name, module, title, module_all_extra):
             # constant, not much fun we can do here except to list it.
             # TODO, figure out some way to document these!
             fw(".. data:: %s\n\n" % attribute)
-            write_indented_lines("   ", fw, "Constant value %s" % repr(value), False)
+            write_indented_lines("   ", fw, "constant value %s" % repr(value), False)
             fw("\n")
         else:
             BPY_LOGGER.debug("\tnot documenting %s.%s of %r type" % (module_name, attribute, value_type.__name__))
@@ -1036,6 +1029,7 @@ def pymodule2sphinx(basepath, module_name, module, title, module_all_extra):
 context_type_map = {
     # context_member: (RNA type, is_collection)
     "active_annotation_layer": ("GPencilLayer", False),
+    "active_base": ("ObjectBase", False),
     "active_bone": ("EditBone", False),
     "active_gpencil_frame": ("GreasePencilLayer", True),
     "active_gpencil_layer": ("GPencilLayer", True),
@@ -1246,7 +1240,7 @@ def pyrna_enum2sphinx(prop, use_empty_descriptions=False):
             "%s.\n" % (
                 identifier,
                 # Account for multi-line enum descriptions, allowing this to be a block of text.
-                indent(" -- ".join(escape_rst(val) for val in (name, description) if val) or "Undocumented", "  "),
+                indent(", ".join(escape_rst(val) for val in (name, description) if val) or "Undocumented", "  "),
             )
             for identifier, name, description in prop.enum_items
         ])
@@ -1555,8 +1549,8 @@ def pyrna2sphinx(basepath):
             fw(".. hlist::\n")
             fw("   :columns: 2\n\n")
 
-            # Context does its own thing.
-            # "active_object": ("Object", False),
+            # context does its own thing
+            # "active_base": ("ObjectBase", False),
             for ref_attr, (ref_type, ref_is_seq) in sorted(context_type_map.items()):
                 if ref_type == struct_id:
                     fw("   * :mod:`bpy.context.%s`\n" % ref_attr)

doc/python_api/static/css/theme_overrides.css

@@ -1,8 +1,7 @@
 /* T76453: Prevent Long enum lists */
-.field-list > dd p {
+.field-list li {
    max-height: 245px;
    overflow-y: auto !important;
-   word-break: break-word;
 }
 
 /* Hide home icon in search area */
@@ -12,15 +11,3 @@
 .wy-nav-content {
     max-width: 1000px !important;
 }
-
-/* Fix long titles on mobile */
-h1, h2, h3, h4, h5, h6 {word-break: break-all}
-
-/* Temp fix for https://github.com/readthedocs/sphinx_rtd_theme/pull/1109 */
-.hlist tr {
-   display: -ms-flexbox;
-   display: flex;
-   flex-flow: row wrap;
- }
-
-.hlist td {margin-right: auto}

extern/audaspace/plugins/ffmpeg/FFMPEGWriter.cpp

@@ -75,7 +75,6 @@ void FFMPEGWriter::encode()
 	m_frame->nb_samples = m_input_samples;
 	m_frame->format = m_codecCtx->sample_fmt;
 	m_frame->channel_layout = m_codecCtx->channel_layout;
-	m_frame->channels = m_specs.channels;
 
 	if(avcodec_fill_audio_frame(m_frame, m_specs.channels, m_codecCtx->sample_fmt, reinterpret_cast<data_t*>(data), m_input_buffer.getSize(), 0) < 0)
 		AUD_THROW(FileException, "File couldn't be written, filling the audio frame failed with ffmpeg.");

extern/mantaflow/preprocessed/fileio/iovdb.cpp

@@ -29,10 +29,10 @@
 
 #if OPENVDB == 1
 #  include "openvdb/openvdb.h"
-#  include "openvdb/points/PointConversion.h"
-#  include "openvdb/points/PointCount.h"
-#  include "openvdb/tools/Clip.h"
-#  include "openvdb/tools/Dense.h"
+#  include <openvdb/points/PointConversion.h>
+#  include <openvdb/points/PointCount.h>
+#  include <openvdb/tools/Clip.h>
+#  include <openvdb/tools/Dense.h>
 #endif
 
 #define POSITION_NAME "P"
@@ -519,7 +519,7 @@ int writeObjectsVDB(const string &filename,
     }
   }
 
-  // Write only if there is at least one grid, optionally write with compression.
+  // Write only if the is at least one grid, optionally write with compression.
   if (gridsVDB.size()) {
     int vdb_flags = openvdb::io::COMPRESS_ACTIVE_MASK;
     switch (compression) {
@@ -534,8 +534,7 @@ int writeObjectsVDB(const string &filename,
       }
       case COMPRESSION_BLOSC: {
 #  if OPENVDB_BLOSC == 1
-        // Cannot use |= here, causes segfault with blosc 1.5.0 (== recommended version)
-        vdb_flags = openvdb::io::COMPRESS_BLOSC;
+        vdb_flags |= openvdb::io::COMPRESS_BLOSC;
 #  else
         debMsg("OpenVDB was built without Blosc support, using Zip compression instead", 1);
         vdb_flags |= openvdb::io::COMPRESS_ZIP;

extern/mantaflow/preprocessed/fluidsolver.h

@@ -384,7 +384,6 @@ class FluidSolver : public PbClass {
   GridStorage<Real> mGrids4dReal;
   GridStorage<Vec3> mGrids4dVec;
   GridStorage<Vec4> mGrids4dVec4;
-
  public:
   PbArgs _args;
 }

extern/mantaflow/preprocessed/general.h

@@ -42,7 +42,7 @@ inline void updateQtGui(bool full, int frame, float time, const std::string &cur
 #  ifdef _DEBUG
 #    define DEBUG 1
 #  endif  // _DEBUG
-#endif    // DEBUG
+#endif  // DEBUG
 
 // Standard exception
 class Error : public std::exception {

extern/mantaflow/preprocessed/gitinfo.h

@@ -1,3 +1,3 @@
 
 
-#define MANTA_GIT_VERSION "commit 9c505cd22e289b98c9aa717efba8ef3201c7e458"
+#define MANTA_GIT_VERSION "commit 39b7a415721ecbf6643612a24e8eadd221aeb934"

extern/mantaflow/preprocessed/grid.h

@@ -389,7 +389,6 @@ class GridBase : public PbClass {
   Real mDx;
   bool m3D;  // precomputed Z shift: to ensure 2D compatibility, always use this instead of sx*sy !
   IndexInt mStrideZ;
-
  public:
   PbArgs _args;
 }

extern/mantaflow/preprocessed/grid4d.h

@@ -326,7 +326,6 @@ class Grid4dBase : public PbClass {
   // precomputed Z,T shift: to ensure 2D compatibility, always use this instead of sx*sy !
   IndexInt mStrideZ;
   IndexInt mStrideT;
-
  public:
   PbArgs _args;
 }
@@ -951,7 +950,6 @@ template<class T> class Grid4d : public Grid4dBase {
 
  protected:
   T *mData;
-
  public:
   PbArgs _args;
 }

extern/mantaflow/preprocessed/levelset.h

@@ -266,7 +266,6 @@ class LevelsetGrid : public Grid<Real> {
   }
 
   static Real invalidTimeValue();
-
  public:
   PbArgs _args;
 }

extern/mantaflow/preprocessed/mesh.h

@@ -796,7 +796,6 @@ class Mesh : public PbClass {
   std::vector<MeshDataImpl<int> *>
       mMdataInt;  //! indicate that mdata of this mesh is copied, and needs to be freed
   bool mFreeMdata;
-
  public:
   PbArgs _args;
 }
@@ -882,7 +881,6 @@ class MeshDataBase : public PbClass {
 
  protected:
   Mesh *mMesh;
-
  public:
   PbArgs _args;
 }
@@ -1647,7 +1645,6 @@ template<class T> class MeshDataImpl : public MeshDataBase {
   //! optionally , we might have an associated grid from which to grab new data
   Grid<T> *mpGridSource;  //! unfortunately , we need to distinguish mac vs regular vec3
   bool mGridSourceMAC;
-
  public:
   PbArgs _args;
 }

extern/mantaflow/preprocessed/movingobs.h

@@ -154,7 +154,6 @@ class MovingObstacle : public PbClass {
   int mEmptyType;
   int mID;
   static int sIDcnt;
-
  public:
   PbArgs _args;
 }

extern/mantaflow/preprocessed/noisefield.h

@@ -236,7 +236,6 @@ class WaveletNoiseField : public PbClass {
   static int randomSeed;
   // global reference count for noise tile
   static std::atomic<int> mNoiseReferenceCount;
-
  public:
   PbArgs _args;
 }

extern/mantaflow/preprocessed/particle.h

@@ -205,7 +205,6 @@ class ParticleBase : public PbClass {
   //! custom seed for particle systems, used by plugins
   int mSeed;  //! fix global random seed storage, used mainly by functions in this class
   static int globalSeed;
-
  public:
   PbArgs _args;
 }
@@ -629,7 +628,6 @@ template<class S> class ParticleSystem : public ParticleBase {
   std::vector<S> mData;
   //! reduce storage , called by doCompress
   virtual void compress();
-
  public:
   PbArgs _args;
 }
@@ -920,7 +918,6 @@ class ParticleIndexSystem : public ParticleSystem<ParticleIndexData> {
       return -1;
     }
   };
-
  public:
   PbArgs _args;
 }
@@ -985,7 +982,6 @@ template<class DATA, class CON> class ConnectedParticleSystem : public ParticleS
  protected:
   std::vector<CON> mSegments;
   virtual void compress();
-
  public:
   PbArgs _args;
 }
@@ -1075,7 +1071,6 @@ class ParticleDataBase : public PbClass {
 
  protected:
   ParticleBase *mpParticleSys;
-
  public:
   PbArgs _args;
 }
@@ -1848,7 +1843,6 @@ template<class T> class ParticleDataImpl : public ParticleDataBase {
   //! optionally , we might have an associated grid from which to grab new data
   Grid<T> *mpGridSource;  //! unfortunately , we need to distinguish mac vs regular vec3
   bool mGridSourceMAC;
-
  public:
   PbArgs _args;
 }

extern/mantaflow/preprocessed/plugin/meshplugins.cpp

@@ -234,10 +234,10 @@ void subdivideMesh(
       normalize(ne2);
 
       // Real thisArea = sqrMag(cross(-e2,e0));
-      //  small angle approximation says sin(x) = arcsin(x) = x,
-      //  arccos(x) = pi/2 - arcsin(x),
-      //  cos(x) = dot(A,B),
-      //  so angle is approximately 1 - dot(A,B).
+      // small angle approximation says sin(x) = arcsin(x) = x,
+      // arccos(x) = pi/2 - arcsin(x),
+      // cos(x) = dot(A,B),
+      // so angle is approximately 1 - dot(A,B).
       Real angle[3];
       angle[0] = 1.0 - dot(ne0, -ne2);
       angle[1] = 1.0 - dot(ne1, -ne0);

extern/mantaflow/preprocessed/plugin/secondaryparticles.cpp

@@ -2287,10 +2287,9 @@ struct knFlipComputePotentialTrappedAir : public KernelBase {
           const Vec3 &vj = scaleFromManta * v.getCentered(x, y, z);
           const Vec3 xij = xi - xj;
           const Vec3 vij = vi - vj;
-          Real h = !pot.is3D() ?
-                       1.414 * radius :
-                       1.732 * radius;  // estimate sqrt(2)*radius resp. sqrt(3)*radius for h, due
-                                        // to squared resp. cubic neighbor area
+          Real h = !pot.is3D() ? 1.414 * radius :
+                                 1.732 * radius;  // estimate sqrt(2)*radius resp. sqrt(3)*radius
+                                                  // for h, due to squared resp. cubic neighbor area
           vdiff += norm(vij) * (1 - dot(getNormalized(vij), getNormalized(xij))) *
                    (1 - norm(xij) / h);
         }

extern/mantaflow/preprocessed/shapes.h

@@ -269,7 +269,6 @@ class Shape : public PbClass {
 
  protected:
   GridType mType;
-
  public:
   PbArgs _args;
 }
@@ -320,7 +319,6 @@ class NullShape : public Shape {
   {
     gridSetConst<Real>(phi, 1000.0f);
   }
-
  public:
   PbArgs _args;
 }
@@ -396,7 +394,6 @@ class Box : public Shape {
 
  protected:
   Vec3 mP0, mP1;
-
  public:
   PbArgs _args;
 }
@@ -458,7 +455,6 @@ class Sphere : public Shape {
  protected:
   Vec3 mCenter, mScale;
   Real mRadius;
-
  public:
   PbArgs _args;
 }
@@ -583,7 +579,6 @@ class Cylinder : public Shape {
  protected:
   Vec3 mCenter, mZDir;
   Real mRadius, mZ;
-
  public:
   PbArgs _args;
 }
@@ -660,7 +655,6 @@ class Slope : public Shape {
   Real mAnglexy, mAngleyz;
   Real mOrigin;
   Vec3 mGs;
-
  public:
   PbArgs _args;
 }

extern/mantaflow/preprocessed/turbulencepart.h

@@ -199,7 +199,6 @@ class TurbulenceParticleSystem : public ParticleSystem<TurbulenceParticleData> {
 
  private:
   WaveletNoiseField &noise;
-
  public:
   PbArgs _args;
 }

extern/mantaflow/preprocessed/vortexpart.h

@@ -127,7 +127,6 @@ class VortexParticleSystem : public ParticleSystem<VortexParticleData> {
   }
 
   virtual ParticleBase *clone();
-
  public:
   PbArgs _args;
 }

extern/mantaflow/preprocessed/vortexsheet.h

@@ -240,7 +240,6 @@ class VortexSheetMesh : public Mesh {
   VorticityChannel mVorticity;
   TexCoord3Channel mTex1, mTex2;
   TurbulenceChannel mTurb;
-
  public:
   PbArgs _args;
 }

extern/smaa_areatex/smaa_areatex.cpp

@@ -1088,8 +1088,6 @@ static int generate_file(AreaOrtho *ortho, AreaDiag *diag, const char *path, boo
 		return 1;
 	}
 
-	// fprintf(stderr, "Generating %s\n", path);
-
 	if (tga)
 		write_tga(ortho, diag, fp, subsampling);
 	else if (raw)

intern/clog/CLG_log.h

@@ -118,7 +118,6 @@ typedef struct CLG_LogType {
 typedef struct CLG_LogRef {
   const char *identifier;
   CLG_LogType *type;
-  struct CLG_LogRef *next;
 } CLG_LogRef;
 
 void CLG_log_str(CLG_LogType *lg,

intern/clog/clog.c

@@ -81,8 +81,6 @@ typedef struct CLG_IDFilter {
 typedef struct CLogContext {
   /** Single linked list of types.  */
   CLG_LogType *types;
-  /** Single linked list of references.  */
-  CLG_LogRef *refs;
 #ifdef WITH_CLOG_PTHREADS
   pthread_mutex_t types_lock;
 #endif
@@ -322,9 +320,7 @@ static bool clg_ctx_filter_check(CLogContext *ctx, const char *identifier)
       if (flt->match[0] == '*' && flt->match[len - 1] == '*') {
         char *match = MEM_callocN(sizeof(char) * len - 1, __func__);
         memcpy(match, flt->match + 1, len - 2);
-        const bool success = (strstr(identifier, match) != NULL);
-        MEM_freeN(match);
-        if (success) {
+        if (strstr(identifier, match) != NULL) {
           return (bool)i;
         }
       }
@@ -677,12 +673,6 @@ static void CLG_ctx_free(CLogContext *ctx)
     MEM_freeN(item);
   }
 
-  while (ctx->refs != NULL) {
-    CLG_LogRef *item = ctx->refs;
-    ctx->refs = item->next;
-    item->type = NULL;
-  }
-
   for (uint i = 0; i < 2; i++) {
     while (ctx->filters[i] != NULL) {
       CLG_IDFilter *item = ctx->filters[i];
@@ -779,10 +769,6 @@ void CLG_logref_init(CLG_LogRef *clg_ref)
   pthread_mutex_lock(&g_ctx->types_lock);
 #endif
   if (clg_ref->type == NULL) {
-    /* Add to the refs list so we can NULL the pointers to 'type' when CLG_exit() is called. */
-    clg_ref->next = g_ctx->refs;
-    g_ctx->refs = clg_ref;
-
     CLG_LogType *clg_ty = clg_ctx_type_find_by_name(g_ctx, clg_ref->identifier);
     if (clg_ty == NULL) {
       clg_ty = clg_ctx_type_register(g_ctx, clg_ref->identifier);

intern/cycles/app/CMakeLists.txt

@@ -71,16 +71,6 @@ if(WITH_CYCLES_STANDALONE)
   target_link_libraries(cycles ${LIBRARIES})
   cycles_target_link_libraries(cycles)
 
-  if(APPLE)
-    if(WITH_OPENCOLORIO)
-      set_property(TARGET cycles APPEND_STRING PROPERTY LINK_FLAGS " -framework IOKit")
-    endif()
-    if(WITH_OPENIMAGEDENOISE AND "${CMAKE_OSX_ARCHITECTURES}" STREQUAL "arm64")
-      # OpenImageDenoise uses BNNS from the Accelerate framework.
-      set_property(TARGET cycles APPEND_STRING PROPERTY LINK_FLAGS " -framework Accelerate")
-    endif()
-  endif()
-
   if(UNIX AND NOT APPLE)
     set_target_properties(cycles PROPERTIES INSTALL_RPATH $ORIGIN/lib)
   endif()

intern/cycles/blender/addon/engine.py

@@ -19,16 +19,16 @@ from __future__ import annotations
 
 
 def _is_using_buggy_driver():
-    import gpu
+    import bgl
     # We need to be conservative here because in multi-GPU systems display card
     # might be quite old, but others one might be just good.
     #
     # So We shouldn't disable possible good dedicated cards just because display
     # card seems weak. And instead we only blacklist configurations which are
     # proven to cause problems.
-    if gpu.platform.vendor_get() == "ATI Technologies Inc.":
+    if bgl.glGetString(bgl.GL_VENDOR) == "ATI Technologies Inc.":
         import re
-        version = gpu.platform.version_get()
+        version = bgl.glGetString(bgl.GL_VERSION)
         if version.endswith("Compatibility Profile Context"):
             # Old HD 4xxx and 5xxx series drivers did not have driver version
             # in the version string, but those cards do not quite work and

intern/cycles/blender/addon/ui.py

@@ -552,9 +552,7 @@ class CYCLES_RENDER_PT_light_paths_fast_gi(CyclesButtonsPanel, Panel):
 
         if world:
           light = world.light_settings
-          col = layout.column(align=True)
-          col.prop(light, "ao_factor", text="AO Factor")
-          col.prop(light, "distance", text="AO Distance")
+          layout.prop(light, "distance", text="AO Distance")
 
 
 class CYCLES_RENDER_PT_motion_blur(CyclesButtonsPanel, Panel):
@@ -725,7 +723,7 @@ class CYCLES_RENDER_PT_performance_tiles(CyclesButtonsPanel, Panel):
         col.prop(cscene, "tile_order", text="Order")
 
         sub = col.column()
-        sub.active = not rd.use_save_buffers and not cscene.use_adaptive_sampling
+        sub.active = not rd.use_save_buffers
         sub.prop(cscene, "use_progressive_refine")
 
 

intern/cycles/blender/blender_camera.cpp

@@ -83,8 +83,6 @@ struct BlenderCamera {
   BoundBox2D pano_viewplane;
   BoundBox2D viewport_camera_border;
 
-  float passepartout_alpha;
-
   Transform matrix;
 
   float offscreen_dicing_scale;
@@ -127,7 +125,6 @@ static void blender_camera_init(BlenderCamera *bcam, BL::RenderSettings &b_rende
   bcam->pano_viewplane.top = 1.0f;
   bcam->viewport_camera_border.right = 1.0f;
   bcam->viewport_camera_border.top = 1.0f;
-  bcam->passepartout_alpha = 0.5f;
   bcam->offscreen_dicing_scale = 1.0f;
   bcam->matrix = transform_identity();
 
@@ -215,8 +212,6 @@ static void blender_camera_from_object(BlenderCamera *bcam,
 
     bcam->lens = b_camera.lens();
 
-    bcam->passepartout_alpha = b_camera.show_passepartout() ? b_camera.passepartout_alpha() : 0.0f;
-
     if (b_camera.dof().use_dof()) {
       /* allow f/stop number to change aperture_size but still
        * give manual control over aperture radius */
@@ -839,19 +834,15 @@ static void blender_camera_border(BlenderCamera *bcam,
                                full_border,
                                &bcam->viewport_camera_border);
 
-  if (b_render.use_border()) {
-    bcam->border.left = b_render.border_min_x();
-    bcam->border.right = b_render.border_max_x();
-    bcam->border.bottom = b_render.border_min_y();
-    bcam->border.top = b_render.border_max_y();
-  }
-  else if (bcam->passepartout_alpha == 1.0f) {
-    bcam->border = full_border;
-  }
-  else {
+  if (!b_render.use_border()) {
     return;
   }
 
+  bcam->border.left = b_render.border_min_x();
+  bcam->border.right = b_render.border_max_x();
+  bcam->border.bottom = b_render.border_min_y();
+  bcam->border.top = b_render.border_max_y();
+
   /* Determine viewport subset matching camera border. */
   blender_camera_border_subset(b_engine,
                                b_render,
@@ -894,7 +885,8 @@ void BlenderSync::sync_view(BL::SpaceView3D &b_v3d,
   }
 }
 
-BufferParams BlenderSync::get_buffer_params(BL::SpaceView3D &b_v3d,
+BufferParams BlenderSync::get_buffer_params(BL::RenderSettings &b_render,
+                                            BL::SpaceView3D &b_v3d,
                                             BL::RegionView3D &b_rv3d,
                                             Camera *cam,
                                             int width,
@@ -910,8 +902,7 @@ BufferParams BlenderSync::get_buffer_params(BL::SpaceView3D &b_v3d,
   if (b_v3d && b_rv3d && b_rv3d.view_perspective() != BL::RegionView3D::view_perspective_CAMERA)
     use_border = b_v3d.use_render_border();
   else
-    /* the camera can always have a passepartout */
-    use_border = true;
+    use_border = b_render.use_border();
 
   if (use_border) {
     /* border render */

intern/cycles/blender/blender_object.cpp

@@ -96,49 +96,7 @@ bool BlenderSync::object_is_light(BL::Object &b_ob)
   return (b_ob_data && b_ob_data.is_a(&RNA_Light));
 }
 
-void BlenderSync::sync_object_motion_init(BL::Object &b_parent, BL::Object &b_ob, Object *object)
-{
-  /* Initialize motion blur for object, detecting if it's enabled and creating motion
-   * steps array if so. */
-  array<Transform> motion;
-  object->set_motion(motion);
-
-  Scene::MotionType need_motion = scene->need_motion();
-  if (need_motion == Scene::MOTION_NONE || !object->get_geometry()) {
-    return;
-  }
-
-  Geometry *geom = object->get_geometry();
-  geom->set_use_motion_blur(false);
-  geom->set_motion_steps(0);
-
-  uint motion_steps;
-
-  if (need_motion == Scene::MOTION_BLUR) {
-    motion_steps = object_motion_steps(b_parent, b_ob, Object::MAX_MOTION_STEPS);
-    geom->set_motion_steps(motion_steps);
-    if (motion_steps && object_use_deform_motion(b_parent, b_ob)) {
-      geom->set_use_motion_blur(true);
-    }
-  }
-  else {
-    motion_steps = 3;
-    geom->set_motion_steps(motion_steps);
-  }
-
-  motion.resize(motion_steps, transform_empty());
-
-  if (motion_steps) {
-    motion[motion_steps / 2] = object->get_tfm();
-
-    /* update motion socket before trying to access object->motion_time */
-    object->set_motion(motion);
-
-    for (size_t step = 0; step < motion_steps; step++) {
-      motion_times.insert(object->motion_time(step));
-    }
-  }
-}
+/* Object */
 
 Object *BlenderSync::sync_object(BL::Depsgraph &b_depsgraph,
                                  BL::ViewLayer &b_view_layer,
@@ -261,8 +219,10 @@ Object *BlenderSync::sync_object(BL::Depsgraph &b_depsgraph,
   }
 
   /* test if we need to sync */
-  bool object_updated = object_map.add_or_update(&object, b_ob, b_parent, key) ||
-                        (tfm != object->get_tfm());
+  bool object_updated = false;
+
+  if (object_map.add_or_update(&object, b_ob, b_parent, key))
+    object_updated = true;
 
   /* mesh sync */
   /* b_ob is owned by the iterator and will go out of scope at the end of the block.
@@ -311,11 +271,49 @@ Object *BlenderSync::sync_object(BL::Depsgraph &b_depsgraph,
    * transform comparison should not be needed, but duplis don't work perfect
    * in the depsgraph and may not signal changes, so this is a workaround */
   if (object->is_modified() || object_updated ||
-      (object->get_geometry() && object->get_geometry()->is_modified())) {
+      (object->get_geometry() && object->get_geometry()->is_modified()) ||
+      tfm != object->get_tfm()) {
     object->name = b_ob.name().c_str();
     object->set_pass_id(b_ob.pass_index());
     object->set_color(get_float3(b_ob.color()));
     object->set_tfm(tfm);
+    array<Transform> motion;
+    object->set_motion(motion);
+
+    /* motion blur */
+    Scene::MotionType need_motion = scene->need_motion();
+    if (need_motion != Scene::MOTION_NONE && object->get_geometry()) {
+      Geometry *geom = object->get_geometry();
+      geom->set_use_motion_blur(false);
+      geom->set_motion_steps(0);
+
+      uint motion_steps;
+
+      if (need_motion == Scene::MOTION_BLUR) {
+        motion_steps = object_motion_steps(b_parent, b_ob, Object::MAX_MOTION_STEPS);
+        geom->set_motion_steps(motion_steps);
+        if (motion_steps && object_use_deform_motion(b_parent, b_ob)) {
+          geom->set_use_motion_blur(true);
+        }
+      }
+      else {
+        motion_steps = 3;
+        geom->set_motion_steps(motion_steps);
+      }
+
+      motion.resize(motion_steps, transform_empty());
+
+      if (motion_steps) {
+        motion[motion_steps / 2] = tfm;
+
+        /* update motion socket before trying to access object->motion_time */
+        object->set_motion(motion);
+
+        for (size_t step = 0; step < motion_steps; step++) {
+          motion_times.insert(object->motion_time(step));
+        }
+      }
+    }
 
     /* dupli texture coordinates and random_id */
     if (is_instance) {
@@ -333,8 +331,6 @@ Object *BlenderSync::sync_object(BL::Depsgraph &b_depsgraph,
     object->tag_update(scene);
   }
 
-  sync_object_motion_init(b_parent, b_ob, object);
-
   if (is_instance) {
     /* Sync possible particle data. */
     sync_dupli_particle(b_parent, b_instance, object);
@@ -564,12 +560,10 @@ void BlenderSync::sync_objects(BL::Depsgraph &b_depsgraph,
   if (!cancel && !motion) {
     sync_background_light(b_v3d, use_portal);
 
-    /* Handle removed data and modified pointers, as this may free memory, delete Nodes in the
-     * right order to ensure that dependent data is freed after their users. Objects should be
-     * freed before particle systems and geometries. */
+    /* handle removed data and modified pointers */
     light_map.post_sync();
-    object_map.post_sync();
     geometry_map.post_sync();
+    object_map.post_sync();
     particle_system_map.post_sync();
   }
 
@@ -617,7 +611,7 @@ void BlenderSync::sync_motion(BL::RenderSettings &b_render,
     if (b_cam) {
       sync_camera_motion(b_render, b_cam, width, height, 0.0f);
     }
-    sync_objects(b_depsgraph, b_v3d);
+    sync_objects(b_depsgraph, b_v3d, 0.0f);
   }
 
   /* Insert motion times from camera. Motion times from other objects

intern/cycles/blender/blender_python.cpp

@@ -35,7 +35,6 @@
 #include "util/util_path.h"
 #include "util/util_string.h"
 #include "util/util_task.h"
-#include "util/util_tbb.h"
 #include "util/util_types.h"
 
 #ifdef WITH_OSL
@@ -289,11 +288,9 @@ static PyObject *render_func(PyObject * /*self*/, PyObject *args)
   RNA_pointer_create(NULL, &RNA_Depsgraph, (ID *)PyLong_AsVoidPtr(pydepsgraph), &depsgraphptr);
   BL::Depsgraph b_depsgraph(depsgraphptr);
 
-  /* Allow Blender to execute other Python scripts, and isolate TBB tasks so we
-   * don't get deadlocks with Blender threads accessing shared data like images. */
   python_thread_state_save(&session->python_thread_state);
 
-  tbb::this_task_arena::isolate([&] { session->render(b_depsgraph); });
+  session->render(b_depsgraph);
 
   python_thread_state_restore(&session->python_thread_state);
 
@@ -330,8 +327,7 @@ static PyObject *bake_func(PyObject * /*self*/, PyObject *args)
 
   python_thread_state_save(&session->python_thread_state);
 
-  tbb::this_task_arena::isolate(
-      [&] { session->bake(b_depsgraph, b_object, pass_type, pass_filter, width, height); });
+  session->bake(b_depsgraph, b_object, pass_type, pass_filter, width, height);
 
   python_thread_state_restore(&session->python_thread_state);
 
@@ -377,7 +373,7 @@ static PyObject *reset_func(PyObject * /*self*/, PyObject *args)
 
   python_thread_state_save(&session->python_thread_state);
 
-  tbb::this_task_arena::isolate([&] { session->reset_session(b_data, b_depsgraph); });
+  session->reset_session(b_data, b_depsgraph);
 
   python_thread_state_restore(&session->python_thread_state);
 
@@ -399,7 +395,7 @@ static PyObject *sync_func(PyObject * /*self*/, PyObject *args)
 
   python_thread_state_save(&session->python_thread_state);
 
-  tbb::this_task_arena::isolate([&] { session->synchronize(b_depsgraph); });
+  session->synchronize(b_depsgraph);
 
   python_thread_state_restore(&session->python_thread_state);
 

intern/cycles/blender/blender_session.cpp

@@ -155,7 +155,7 @@ void BlenderSession::create_session()
 
   /* set buffer parameters */
   BufferParams buffer_params = BlenderSync::get_buffer_params(
-      b_v3d, b_rv3d, scene->camera, width, height, session_params.denoising.use);
+      b_render, b_v3d, b_rv3d, scene->camera, width, height, session_params.denoising.use);
   session->reset(buffer_params, session_params.samples);
 
   b_engine.use_highlight_tiles(session_params.progressive_refine == false);
@@ -237,12 +237,10 @@ void BlenderSession::reset_session(BL::BlendData &b_data, BL::Depsgraph &b_depsg
     sync->sync_recalc(b_depsgraph, b_v3d);
   }
 
-  BL::Object b_camera_override(b_engine.camera_override());
-  sync->sync_camera(b_render, b_camera_override, width, height, "");
-
   BL::SpaceView3D b_null_space_view3d(PointerRNA_NULL);
   BL::RegionView3D b_null_region_view3d(PointerRNA_NULL);
-  BufferParams buffer_params = BlenderSync::get_buffer_params(b_null_space_view3d,
+  BufferParams buffer_params = BlenderSync::get_buffer_params(b_render,
+                                                              b_null_space_view3d,
                                                               b_null_region_view3d,
                                                               scene->camera,
                                                               width,
@@ -485,7 +483,7 @@ void BlenderSession::render(BL::Depsgraph &b_depsgraph_)
   SessionParams session_params = BlenderSync::get_session_params(
       b_engine, b_userpref, b_scene, background, b_view_layer);
   BufferParams buffer_params = BlenderSync::get_buffer_params(
-      b_v3d, b_rv3d, scene->camera, width, height, session_params.denoising.use);
+      b_render, b_v3d, b_rv3d, scene->camera, width, height, session_params.denoising.use);
 
   /* temporary render result to find needed passes and views */
   BL::RenderResult b_rr = begin_render_result(
@@ -809,7 +807,7 @@ void BlenderSession::synchronize(BL::Depsgraph &b_depsgraph_)
 
   /* get buffer parameters */
   BufferParams buffer_params = BlenderSync::get_buffer_params(
-      b_v3d, b_rv3d, scene->camera, width, height, session_params.denoising.use);
+      b_render, b_v3d, b_rv3d, scene->camera, width, height, session_params.denoising.use);
 
   if (!buffer_params.denoising_data_pass) {
     session_params.denoising.use = false;
@@ -888,7 +886,7 @@ bool BlenderSession::draw(int w, int h)
       SessionParams session_params = BlenderSync::get_session_params(
           b_engine, b_userpref, b_scene, background);
       BufferParams buffer_params = BlenderSync::get_buffer_params(
-          b_v3d, b_rv3d, scene->camera, width, height, session_params.denoising.use);
+          b_render, b_v3d, b_rv3d, scene->camera, width, height, session_params.denoising.use);
       bool session_pause = BlenderSync::get_session_pause(b_scene, background);
 
       if (session_pause == false) {
@@ -906,7 +904,7 @@ bool BlenderSession::draw(int w, int h)
 
   /* draw */
   BufferParams buffer_params = BlenderSync::get_buffer_params(
-      b_v3d, b_rv3d, scene->camera, width, height, session->params.denoising.use);
+      b_render, b_v3d, b_rv3d, scene->camera, width, height, session->params.denoising.use);
   DeviceDrawParams draw_params;
 
   if (session->params.display_buffer_linear) {

intern/cycles/blender/blender_shader.cpp

@@ -1373,7 +1373,7 @@ void BlenderSync::sync_world(BL::Depsgraph &b_depsgraph, BL::SpaceView3D &b_v3d,
   BlenderViewportParameters new_viewport_parameters(b_v3d);
 
   if (world_recalc || update_all || b_world.ptr.data != world_map ||
-      viewport_parameters.shader_modified(new_viewport_parameters)) {
+      viewport_parameters.modified(new_viewport_parameters)) {
     Shader *shader = scene->default_background;
     ShaderGraph *graph = new ShaderGraph();
 
@@ -1501,8 +1501,8 @@ void BlenderSync::sync_world(BL::Depsgraph &b_depsgraph, BL::SpaceView3D &b_v3d,
     background->set_transparent_roughness_threshold(0.0f);
   }
 
-  background->set_use_shader(view_layer.use_background_shader ||
-                             viewport_parameters.use_custom_shader());
+  background->set_use_shader(view_layer.use_background_shader |
+                             viewport_parameters.custom_viewport_parameters());
   background->set_use_ao(background->get_use_ao() && view_layer.use_background_ao);
 
   background->tag_update(scene);
@@ -1553,9 +1553,13 @@ void BlenderSync::sync_lights(BL::Depsgraph &b_depsgraph, bool update_all)
 void BlenderSync::sync_shaders(BL::Depsgraph &b_depsgraph, BL::SpaceView3D &b_v3d)
 {
   /* for auto refresh images */
-  ImageManager *image_manager = scene->image_manager;
-  const int frame = b_scene.frame_current();
-  const bool auto_refresh_update = image_manager->set_animation_frame_update(frame);
+  bool auto_refresh_update = false;
+
+  if (preview) {
+    ImageManager *image_manager = scene->image_manager;
+    int frame = b_scene.frame_current();
+    auto_refresh_update = image_manager->set_animation_frame_update(frame);
+  }
 
   shader_map.pre_sync();
 

intern/cycles/blender/blender_sync.cpp

@@ -69,8 +69,7 @@ BlenderSync::BlenderSync(BL::RenderEngine &b_engine,
       experimental(false),
       dicing_rate(1.0f),
       max_subdivisions(12),
-      progress(progress),
-      has_updates_(true)
+      progress(progress)
 {
   PointerRNA cscene = RNA_pointer_get(&b_scene.ptr, "cycles");
   dicing_rate = preview ? RNA_float_get(&cscene, "preview_dicing_rate") :
@@ -85,9 +84,7 @@ BlenderSync::~BlenderSync()
 void BlenderSync::reset(BL::BlendData &b_data, BL::Scene &b_scene)
 {
   /* Update data and scene pointers in case they change in session reset,
-   * for example after undo.
-   * Note that we do not modify the `has_updates_` flag here because the sync
-   * reset is also used during viewport navigation. */
+   * for example after undo. */
   this->b_data = b_data;
   this->b_scene = b_scene;
 }
@@ -120,8 +117,6 @@ void BlenderSync::sync_recalc(BL::Depsgraph &b_depsgraph, BL::SpaceView3D &b_v3d
     }
 
     if (dicing_prop_changed) {
-      has_updates_ = true;
-
       for (const pair<const GeometryKey, Geometry *> &iter : geometry_map.key_to_scene_data()) {
         Geometry *geom = iter.second;
         if (geom->is_mesh()) {
@@ -138,12 +133,6 @@ void BlenderSync::sync_recalc(BL::Depsgraph &b_depsgraph, BL::SpaceView3D &b_v3d
 
   /* Iterate over all IDs in this depsgraph. */
   for (BL::DepsgraphUpdate &b_update : b_depsgraph.updates) {
-    /* TODO(sergey): Can do more selective filter here. For example, ignore changes made to
-     * screen datablock. Note that sync_data() needs to be called after object deletion, and
-     * currently this is ensured by the scene ID tagged for update, which sets the `has_updates_`
-     * flag. */
-    has_updates_ = true;
-
     BL::ID b_id(b_update.id());
 
     /* Material */
@@ -224,13 +213,9 @@ void BlenderSync::sync_recalc(BL::Depsgraph &b_depsgraph, BL::SpaceView3D &b_v3d
 
   if (b_v3d) {
     BlenderViewportParameters new_viewport_parameters(b_v3d);
-
-    if (viewport_parameters.shader_modified(new_viewport_parameters)) {
+    if (viewport_parameters.modified(new_viewport_parameters)) {
       world_recalc = true;
-      has_updates_ = true;
     }
-
-    has_updates_ |= viewport_parameters.modified(new_viewport_parameters);
   }
 }
 
@@ -242,15 +227,11 @@ void BlenderSync::sync_data(BL::RenderSettings &b_render,
                             int height,
                             void **python_thread_state)
 {
-  if (!has_updates_) {
-    return;
-  }
-
   scoped_timer timer;
 
   BL::ViewLayer b_view_layer = b_depsgraph.view_layer_eval();
 
-  sync_view_layer(b_view_layer);
+  sync_view_layer(b_v3d, b_view_layer);
   sync_integrator();
   sync_film(b_v3d);
   sync_shaders(b_depsgraph, b_v3d);
@@ -273,8 +254,6 @@ void BlenderSync::sync_data(BL::RenderSettings &b_render,
   free_data_after_sync(b_depsgraph);
 
   VLOG(1) << "Total time spent synchronizing data: " << timer.get_time();
-
-  has_updates_ = false;
 }
 
 /* Integrator */
@@ -445,7 +424,7 @@ void BlenderSync::sync_film(BL::SpaceView3D &b_v3d)
 
 /* Render Layer */
 
-void BlenderSync::sync_view_layer(BL::ViewLayer &b_view_layer)
+void BlenderSync::sync_view_layer(BL::SpaceView3D & /*b_v3d*/, BL::ViewLayer &b_view_layer)
 {
   view_layer.name = b_view_layer.name();
 
@@ -760,18 +739,12 @@ void BlenderSync::free_data_after_sync(BL::Depsgraph &b_depsgraph)
    * caches to be releases from blender side in order to reduce peak memory
    * footprint during synchronization process.
    */
-
   const bool is_interface_locked = b_engine.render() && b_engine.render().use_lock_interface();
-  const bool is_persistent_data = b_engine.render() && b_engine.render().use_persistent_data();
-  const bool can_free_caches =
-      (BlenderSession::headless || is_interface_locked) &&
-      /* Baking re-uses the depsgraph multiple times, clearing crashes
-       * reading un-evaluated mesh data which isn't aligned with the
-       * geometry we're baking, see T71012. */
-      !scene->bake_manager->get_baking() &&
-      /* Persistent data must main caches for performance and correctness. */
-      !is_persistent_data;
-
+  const bool can_free_caches = (BlenderSession::headless || is_interface_locked) &&
+                               /* Baking re-uses the depsgraph multiple times, clearing crashes
+                                * reading un-evaluated mesh data which isn't aligned with the
+                                * geometry we're baking, see T71012. */
+                               !scene->bake_manager->get_baking();
   if (!can_free_caches) {
     return;
   }
@@ -896,9 +869,6 @@ SessionParams BlenderSync::get_session_params(BL::RenderEngine &b_engine,
   /* Clamp samples. */
   params.samples = min(params.samples, Integrator::MAX_SAMPLES);
 
-  /* Adaptive sampling. */
-  params.adaptive_sampling = RNA_boolean_get(&cscene, "use_adaptive_sampling");
-
   /* tiles */
   const bool is_cpu = (params.device.type == DEVICE_CPU);
   if (!is_cpu && !background) {
@@ -951,7 +921,7 @@ SessionParams BlenderSync::get_session_params(BL::RenderEngine &b_engine,
   BL::RenderSettings b_r = b_scene.render();
   params.progressive_refine = b_engine.is_preview() ||
                               get_boolean(cscene, "use_progressive_refine");
-  if (b_r.use_save_buffers() || params.adaptive_sampling)
+  if (b_r.use_save_buffers())
     params.progressive_refine = false;
 
   if (background) {
@@ -987,6 +957,8 @@ SessionParams BlenderSync::get_session_params(BL::RenderEngine &b_engine,
   params.use_profiling = params.device.has_profiling && !b_engine.is_preview() && background &&
                          BlenderSession::print_render_stats;
 
+  params.adaptive_sampling = RNA_boolean_get(&cscene, "use_adaptive_sampling");
+
   return params;
 }
 

intern/cycles/blender/blender_sync.h

@@ -73,7 +73,7 @@ class BlenderSync {
                  int width,
                  int height,
                  void **python_thread_state);
-  void sync_view_layer(BL::ViewLayer &b_view_layer);
+  void sync_view_layer(BL::SpaceView3D &b_v3d, BL::ViewLayer &b_view_layer);
   vector<Pass> sync_render_passes(BL::Scene &b_scene,
                                   BL::RenderLayer &b_render_layer,
                                   BL::ViewLayer &b_view_layer,
@@ -104,7 +104,8 @@ class BlenderSync {
       bool background,
       BL::ViewLayer b_view_layer = BL::ViewLayer(PointerRNA_NULL));
   static bool get_session_pause(BL::Scene &b_scene, bool background);
-  static BufferParams get_buffer_params(BL::SpaceView3D &b_v3d,
+  static BufferParams get_buffer_params(BL::RenderSettings &b_render,
+                                        BL::SpaceView3D &b_v3d,
                                         BL::RegionView3D &b_rv3d,
                                         Camera *cam,
                                         int width,
@@ -149,7 +150,6 @@ class BlenderSync {
                       BlenderObjectCulling &culling,
                       bool *use_portal,
                       TaskPool *geom_task_pool);
-  void sync_object_motion_init(BL::Object &b_parent, BL::Object &b_ob, Object *object);
 
   bool sync_object_attributes(BL::DepsgraphObjectInstance &b_instance, Object *object);
 
@@ -263,12 +263,6 @@ class BlenderSync {
   } view_layer;
 
   Progress &progress;
-
- protected:
-  /* Indicates that `sync_recalc()` detected changes in the scene.
-   * If this flag is false then the data is considered to be up-to-date and will not be
-   * synchronized at all. */
-  bool has_updates_ = true;
 };
 
 CCL_NAMESPACE_END

intern/cycles/blender/blender_viewport.cpp

@@ -13,7 +13,6 @@
  * See the License for the specific language governing permissions and
  * limitations under the License.
  */
-
 #include "blender_viewport.h"
 
 #include "blender_util.h"
@@ -26,39 +25,29 @@ BlenderViewportParameters::BlenderViewportParameters()
       studiolight_rotate_z(0.0f),
       studiolight_intensity(1.0f),
       studiolight_background_alpha(1.0f),
-      display_pass(PASS_COMBINED)
+      studiolight_path(ustring())
 {
 }
 
 BlenderViewportParameters::BlenderViewportParameters(BL::SpaceView3D &b_v3d)
     : BlenderViewportParameters()
 {
-  if (!b_v3d) {
-    return;
-  }
-
-  BL::View3DShading shading = b_v3d.shading();
-  PointerRNA cshading = RNA_pointer_get(&shading.ptr, "cycles");
-
-  /* We only copy the shading parameters if we are in look dev mode. otherwise
+  /* We only copy the parameters if we are in look dev mode. otherwise
    * defaults are being used. These defaults mimic normal render settings */
-  if (shading.type() == BL::View3DShading::type_RENDERED) {
-    use_scene_world = shading.use_scene_world_render();
-    use_scene_lights = shading.use_scene_lights_render();
-
+  if (b_v3d && b_v3d.shading().type() == BL::View3DShading::type_RENDERED) {
+    use_scene_world = b_v3d.shading().use_scene_world_render();
+    use_scene_lights = b_v3d.shading().use_scene_lights_render();
     if (!use_scene_world) {
-      studiolight_rotate_z = shading.studiolight_rotate_z();
-      studiolight_intensity = shading.studiolight_intensity();
-      studiolight_background_alpha = shading.studiolight_background_alpha();
-      studiolight_path = shading.selected_studio_light().path();
+      studiolight_rotate_z = b_v3d.shading().studiolight_rotate_z();
+      studiolight_intensity = b_v3d.shading().studiolight_intensity();
+      studiolight_background_alpha = b_v3d.shading().studiolight_background_alpha();
+      studiolight_path = b_v3d.shading().selected_studio_light().path();
     }
   }
-
-  /* Film. */
-  display_pass = (PassType)get_enum(cshading, "render_pass", -1, -1);
 }
 
-bool BlenderViewportParameters::shader_modified(const BlenderViewportParameters &other) const
+/* Check if two instances are different. */
+const bool BlenderViewportParameters::modified(const BlenderViewportParameters &other) const
 {
   return use_scene_world != other.use_scene_world || use_scene_lights != other.use_scene_lights ||
          studiolight_rotate_z != other.studiolight_rotate_z ||
@@ -67,26 +56,26 @@ bool BlenderViewportParameters::shader_modified(const BlenderViewportParameters
          studiolight_path != other.studiolight_path;
 }
 
-bool BlenderViewportParameters::film_modified(const BlenderViewportParameters &other) const
-{
-  return display_pass != other.display_pass;
-}
-
-bool BlenderViewportParameters::modified(const BlenderViewportParameters &other) const
-{
-  return shader_modified(other) || film_modified(other);
-}
-
-bool BlenderViewportParameters::use_custom_shader() const
+const bool BlenderViewportParameters::custom_viewport_parameters() const
 {
   return !(use_scene_world && use_scene_lights);
 }
 
+PassType BlenderViewportParameters::get_viewport_display_render_pass(BL::SpaceView3D &b_v3d)
+{
+  PassType display_pass = PASS_NONE;
+  if (b_v3d) {
+    BL::View3DShading b_view3dshading = b_v3d.shading();
+    PointerRNA cshading = RNA_pointer_get(&b_view3dshading.ptr, "cycles");
+    display_pass = (PassType)get_enum(cshading, "render_pass", -1, -1);
+  }
+  return display_pass;
+}
+
 PassType update_viewport_display_passes(BL::SpaceView3D &b_v3d, vector<Pass> &passes)
 {
   if (b_v3d) {
-    const BlenderViewportParameters viewport_parameters(b_v3d);
-    const PassType display_pass = viewport_parameters.display_pass;
+    PassType display_pass = BlenderViewportParameters::get_viewport_display_render_pass(b_v3d);
 
     passes.clear();
     Pass::add(display_pass, passes);

intern/cycles/blender/blender_viewport.h

@@ -18,18 +18,17 @@
 #define __BLENDER_VIEWPORT_H__
 
 #include "MEM_guardedalloc.h"
-
 #include "RNA_access.h"
 #include "RNA_blender_cpp.h"
 #include "RNA_types.h"
 
 #include "render/film.h"
+#include "util/util_param.h"
 
 CCL_NAMESPACE_BEGIN
 
 class BlenderViewportParameters {
- public:
-  /* Shader. */
+ private:
   bool use_scene_world;
   bool use_scene_lights;
   float studiolight_rotate_z;
@@ -37,24 +36,17 @@ class BlenderViewportParameters {
   float studiolight_background_alpha;
   ustring studiolight_path;
 
-  /* Film. */
-  PassType display_pass;
-
   BlenderViewportParameters();
-  explicit BlenderViewportParameters(BL::SpaceView3D &b_v3d);
+  BlenderViewportParameters(BL::SpaceView3D &b_v3d);
 
-  /* Check whether any of shading related settings are different from the given parameters. */
-  bool shader_modified(const BlenderViewportParameters &other) const;
+  const bool modified(const BlenderViewportParameters &other) const;
+  const bool custom_viewport_parameters() const;
+  friend class BlenderSync;
 
-  /* Check whether any of film related settings are different from the given parameters. */
-  bool film_modified(const BlenderViewportParameters &other) const;
-
-  /* Check whether any of settings are different from the given parameters. */
-  bool modified(const BlenderViewportParameters &other) const;
-
-  /* Returns truth when a custom shader defined by the viewport is to be used instead of the
-   * regular background shader or scene light. */
-  bool use_custom_shader() const;
+ public:
+  /* Retrieve the render pass that needs to be displayed on the given `SpaceView3D`
+   * When the `b_v3d` parameter is not given `PASS_NONE` will be returned. */
+  static PassType get_viewport_display_render_pass(BL::SpaceView3D &b_v3d);
 };
 
 PassType update_viewport_display_passes(BL::SpaceView3D &b_v3d, vector<Pass> &passes);

intern/cycles/bvh/bvh_optix.cpp

@@ -17,8 +17,6 @@
 
 #ifdef WITH_OPTIX
 
-#  include "device/device.h"
-
 #  include "bvh/bvh_optix.h"
 
 CCL_NAMESPACE_BEGIN
@@ -28,7 +26,6 @@ BVHOptiX::BVHOptiX(const BVHParams &params_,
                    const vector<Object *> &objects_,
                    Device *device)
     : BVH(params_, geometry_, objects_),
-      device(device),
       traversable_handle(0),
       as_data(device, params_.top_level ? "optix tlas" : "optix blas", false),
       motion_transform_data(device, "optix motion transform", false)
@@ -37,9 +34,7 @@ BVHOptiX::BVHOptiX(const BVHParams &params_,
 
 BVHOptiX::~BVHOptiX()
 {
-  // Acceleration structure memory is delayed freed on device, since deleting the
-  // BVH may happen while still being used for rendering.
-  device->release_optix_bvh(this);
+  // Acceleration structure memory is freed via the 'as_data' destructor
 }
 
 CCL_NAMESPACE_END

intern/cycles/bvh/bvh_optix.h

@@ -28,7 +28,6 @@ CCL_NAMESPACE_BEGIN
 
 class BVHOptiX : public BVH {
  public:
-  Device *device;
   uint64_t traversable_handle;
   device_only_memory<char> as_data;
   device_only_memory<char> motion_transform_data;

intern/cycles/device/device.h

@@ -61,6 +61,7 @@ enum DeviceTypeMask {
 };
 
 enum DeviceKernelStatus {
+  DEVICE_KERNEL_WAITING_FOR_FEATURE_KERNEL = 0,
   DEVICE_KERNEL_FEATURE_KERNEL_AVAILABLE,
   DEVICE_KERNEL_USING_FEATURE_KERNEL,
   DEVICE_KERNEL_FEATURE_KERNEL_INVALID,
@@ -426,9 +427,6 @@ class Device {
   /* acceleration structure building */
   virtual void build_bvh(BVH *bvh, Progress &progress, bool refit);
 
-  /* OptiX specific destructor. */
-  virtual void release_optix_bvh(BVH * /*bvh*/){};
-
 #ifdef WITH_NETWORK
   /* networking */
   void server_run();

intern/cycles/device/device_memory.cpp

@@ -35,54 +35,10 @@ device_memory::device_memory(Device *device, const char *name, MemoryType type)
       device_pointer(0),
       host_pointer(0),
       shared_pointer(0),
-      shared_counter(0),
-      original_device_ptr(0),
-      original_device_size(0),
-      original_device(0),
-      need_realloc_(false),
-      modified(false)
+      shared_counter(0)
 {
 }
 
-device_memory::device_memory(device_memory &&other) noexcept
-    : data_type(other.data_type),
-      data_elements(other.data_elements),
-      data_size(other.data_size),
-      device_size(other.device_size),
-      data_width(other.data_width),
-      data_height(other.data_height),
-      data_depth(other.data_depth),
-      type(other.type),
-      name(other.name),
-      device(other.device),
-      device_pointer(other.device_pointer),
-      host_pointer(other.host_pointer),
-      shared_pointer(other.shared_pointer),
-      shared_counter(other.shared_counter),
-      original_device_ptr(other.original_device_ptr),
-      original_device_size(other.original_device_size),
-      original_device(other.original_device),
-      need_realloc_(other.need_realloc_),
-      modified(other.modified)
-{
-  other.data_elements = 0;
-  other.data_size = 0;
-  other.device_size = 0;
-  other.data_width = 0;
-  other.data_height = 0;
-  other.data_depth = 0;
-  other.device = 0;
-  other.device_pointer = 0;
-  other.host_pointer = 0;
-  other.shared_pointer = 0;
-  other.shared_counter = 0;
-  other.original_device_ptr = 0;
-  other.original_device_size = 0;
-  other.original_device = 0;
-  other.need_realloc_ = false;
-  other.modified = false;
-}
-
 device_memory::~device_memory()
 {
   assert(shared_pointer == 0);

intern/cycles/device/device_memory.h

@@ -238,7 +238,6 @@ class device_memory {
 
   /* Only create through subclasses. */
   device_memory(Device *device, const char *name, MemoryType type);
-  device_memory(device_memory &&other) noexcept;
 
   /* No copying allowed. */
   device_memory(const device_memory &) = delete;
@@ -278,10 +277,6 @@ template<typename T> class device_only_memory : public device_memory {
     data_elements = max(device_type_traits<T>::num_elements, 1);
   }
 
-  device_only_memory(device_only_memory &&other) noexcept : device_memory(std::move(other))
-  {
-  }
-
   virtual ~device_only_memory()
   {
     free();

intern/cycles/device/device_multi.cpp

@@ -232,6 +232,10 @@ class MultiDevice : public Device {
     foreach (SubDevice &sub, devices) {
       DeviceKernelStatus subresult = sub.device->get_active_kernel_switch_state();
       switch (subresult) {
+        case DEVICE_KERNEL_WAITING_FOR_FEATURE_KERNEL:
+          result = subresult;
+          break;
+
         case DEVICE_KERNEL_FEATURE_KERNEL_INVALID:
         case DEVICE_KERNEL_FEATURE_KERNEL_AVAILABLE:
           return subresult;

intern/cycles/device/device_optix.cpp

@@ -193,9 +193,6 @@ class OptiXDevice : public CUDADevice {
   device_only_memory<unsigned char> denoiser_state;
   int denoiser_input_passes = 0;
 
-  vector<device_only_memory<char>> delayed_free_bvh_memory;
-  thread_mutex delayed_free_bvh_mutex;
-
  public:
   OptiXDevice(DeviceInfo &info_, Stats &stats_, Profiler &profiler_, bool background_)
       : CUDADevice(info_, stats_, profiler_, background_),
@@ -261,8 +258,6 @@ class OptiXDevice : public CUDADevice {
     // Make CUDA context current
     const CUDAContextScope scope(cuContext);
 
-    free_bvh_memory_delayed();
-
     sbt_data.free();
     texture_info.free();
     launch_params.free();
@@ -726,11 +721,7 @@ class OptiXDevice : public CUDADevice {
       }
     }
     else if (task.type == DeviceTask::SHADER) {
-      // CUDA kernels are used when doing baking
-      if (optix_module == NULL)
-        CUDADevice::shader(task);
-      else
-        launch_shader_eval(task, thread_index);
+      launch_shader_eval(task, thread_index);
     }
     else if (task.type == DeviceTask::DENOISE_BUFFER) {
       // Set up a single tile that covers the whole task and denoise it
@@ -1306,8 +1297,6 @@ class OptiXDevice : public CUDADevice {
       return;
     }
 
-    free_bvh_memory_delayed();
-
     BVHOptiX *const bvh_optix = static_cast<BVHOptiX *>(bvh);
 
     progress.set_substatus("Building OptiX acceleration structure");
@@ -1778,24 +1767,6 @@ class OptiXDevice : public CUDADevice {
     }
   }
 
-  void release_optix_bvh(BVH *bvh) override
-  {
-    thread_scoped_lock lock(delayed_free_bvh_mutex);
-    /* Do delayed free of BVH memory, since geometry holding BVH might be deleted
-     * while GPU is still rendering. */
-    BVHOptiX *const bvh_optix = static_cast<BVHOptiX *>(bvh);
-
-    delayed_free_bvh_memory.emplace_back(std::move(bvh_optix->as_data));
-    delayed_free_bvh_memory.emplace_back(std::move(bvh_optix->motion_transform_data));
-    bvh_optix->traversable_handle = 0;
-  }
-
-  void free_bvh_memory_delayed()
-  {
-    thread_scoped_lock lock(delayed_free_bvh_mutex);
-    delayed_free_bvh_memory.free_memory();
-  }
-
   void const_copy_to(const char *name, void *host, size_t size) override
   {
     // Set constant memory for CUDA module

intern/cycles/device/opencl/device_opencl.h

@@ -269,6 +269,7 @@ class OpenCLDevice : public Device {
   cl_device_id cdDevice;
   cl_int ciErr;
   int device_num;
+  bool use_preview_kernels;
 
   class OpenCLProgram {
    public:
@@ -368,7 +369,8 @@ class OpenCLDevice : public Device {
     /* Load the kernels and put the created kernels in the given
      * `programs` parameter. */
     void load_kernels(vector<OpenCLProgram *> &programs,
-                      const DeviceRequestedFeatures &requested_features);
+                      const DeviceRequestedFeatures &requested_features,
+                      bool is_preview = false);
   };
 
   DeviceSplitKernel *split_kernel;
@@ -380,6 +382,7 @@ class OpenCLDevice : public Device {
   OpenCLProgram denoising_program;
 
   OpenCLSplitPrograms kernel_programs;
+  OpenCLSplitPrograms preview_programs;
 
   typedef map<string, device_vector<uchar> *> ConstMemMap;
   typedef map<string, device_ptr> MemMap;
@@ -409,6 +412,7 @@ class OpenCLDevice : public Device {
   string device_md5_hash(string kernel_custom_build_options = "");
   bool load_kernels(const DeviceRequestedFeatures &requested_features);
   void load_required_kernels(const DeviceRequestedFeatures &requested_features);
+  void load_preview_kernels();
 
   bool wait_for_availability(const DeviceRequestedFeatures &requested_features);
   DeviceKernelStatus get_active_kernel_switch_state();
@@ -418,7 +422,8 @@ class OpenCLDevice : public Device {
   /* Get the program file name to compile (*.cl) for the given kernel */
   const string get_opencl_program_filename(const string &kernel_name);
   string get_build_options(const DeviceRequestedFeatures &requested_features,
-                           const string &opencl_program_name);
+                           const string &opencl_program_name,
+                           bool preview_kernel = false);
   /* Enable the default features to reduce recompilation events */
   void enable_default_features(DeviceRequestedFeatures &features);
 

intern/cycles/device/opencl/device_opencl_impl.cpp

@@ -107,7 +107,8 @@ void OpenCLDevice::enable_default_features(DeviceRequestedFeatures &features)
 }
 
 string OpenCLDevice::get_build_options(const DeviceRequestedFeatures &requested_features,
-                                       const string &opencl_program_name)
+                                       const string &opencl_program_name,
+                                       bool preview_kernel)
 {
   /* first check for non-split kernel programs */
   if (opencl_program_name == "base" || opencl_program_name == "denoising") {
@@ -184,7 +185,13 @@ string OpenCLDevice::get_build_options(const DeviceRequestedFeatures &requested_
   enable_default_features(nofeatures);
 
   /* Add program specific optimized compile directives */
-  if (opencl_program_name == "split_do_volume" && !requested_features.use_volume) {
+  if (preview_kernel) {
+    DeviceRequestedFeatures preview_features;
+    preview_features.use_hair = true;
+    build_options += "-D__KERNEL_AO_PREVIEW__ ";
+    build_options += preview_features.get_build_options();
+  }
+  else if (opencl_program_name == "split_do_volume" && !requested_features.use_volume) {
     build_options += nofeatures.get_build_options();
   }
   else {
@@ -231,7 +238,9 @@ OpenCLDevice::OpenCLSplitPrograms::~OpenCLSplitPrograms()
 }
 
 void OpenCLDevice::OpenCLSplitPrograms::load_kernels(
-    vector<OpenCLProgram *> &programs, const DeviceRequestedFeatures &requested_features)
+    vector<OpenCLProgram *> &programs,
+    const DeviceRequestedFeatures &requested_features,
+    bool is_preview)
 {
   if (!requested_features.use_baking) {
 #  define ADD_SPLIT_KERNEL_BUNDLE_PROGRAM(kernel_name) \
@@ -242,7 +251,7 @@ void OpenCLDevice::OpenCLSplitPrograms::load_kernels(
         device, \
         program_name_##kernel_name, \
         "kernel_" #kernel_name ".cl", \
-        device->get_build_options(requested_features, program_name_##kernel_name)); \
+        device->get_build_options(requested_features, program_name_##kernel_name, is_preview)); \
     program_##kernel_name.add_kernel(ustring("path_trace_" #kernel_name)); \
     programs.push_back(&program_##kernel_name);
 
@@ -250,7 +259,7 @@ void OpenCLDevice::OpenCLSplitPrograms::load_kernels(
     ADD_SPLIT_KERNEL_PROGRAM(subsurface_scatter);
     ADD_SPLIT_KERNEL_PROGRAM(direct_lighting);
     ADD_SPLIT_KERNEL_PROGRAM(indirect_background);
-    if (requested_features.use_volume) {
+    if (requested_features.use_volume || is_preview) {
       ADD_SPLIT_KERNEL_PROGRAM(do_volume);
     }
     ADD_SPLIT_KERNEL_PROGRAM(shader_eval);
@@ -265,7 +274,7 @@ void OpenCLDevice::OpenCLSplitPrograms::load_kernels(
         device,
         "split_bundle",
         "kernel_split_bundle.cl",
-        device->get_build_options(requested_features, "split_bundle"));
+        device->get_build_options(requested_features, "split_bundle", is_preview));
 
     ADD_SPLIT_KERNEL_BUNDLE_PROGRAM(data_init);
     ADD_SPLIT_KERNEL_BUNDLE_PROGRAM(state_buffer_size);
@@ -394,7 +403,7 @@ class OpenCLSplitKernel : public DeviceSplitKernel {
         device,
         program_name,
         device->get_opencl_program_filename(kernel_name),
-        device->get_build_options(requested_features, program_name));
+        device->get_build_options(requested_features, program_name, device->use_preview_kernels));
 
     kernel->program.add_kernel(ustring("path_trace_" + kernel_name));
     kernel->program.load();
@@ -560,11 +569,6 @@ class OpenCLSplitKernel : public DeviceSplitKernel {
     size_t num_elements = max_elements_for_max_buffer_size(kg, data, max_buffer_size);
     int2 global_size = make_int2(max(round_down((int)sqrt(num_elements), 64), 64),
                                  (int)sqrt(num_elements));
-
-    if (device->info.description.find("Intel") != string::npos) {
-      global_size = make_int2(min(512, global_size.x), min(512, global_size.y));
-    }
-
     VLOG(1) << "Global size: " << global_size << ".";
     return global_size;
   }
@@ -608,6 +612,7 @@ OpenCLDevice::OpenCLDevice(DeviceInfo &info, Stats &stats, Profiler &profiler, b
     : Device(info, stats, profiler, background),
       load_kernel_num_compiling(0),
       kernel_programs(this),
+      preview_programs(this),
       memory_manager(this),
       texture_info(this, "__texture_info", MEM_GLOBAL)
 {
@@ -617,6 +622,7 @@ OpenCLDevice::OpenCLDevice(DeviceInfo &info, Stats &stats, Profiler &profiler, b
   cqCommandQueue = NULL;
   device_initialized = false;
   textures_need_update = true;
+  use_preview_kernels = !background;
 
   vector<OpenCLPlatformDevice> usable_devices;
   OpenCLInfo::get_usable_devices(&usable_devices);
@@ -672,6 +678,9 @@ OpenCLDevice::OpenCLDevice(DeviceInfo &info, Stats &stats, Profiler &profiler, b
   device_initialized = true;
 
   split_kernel = new OpenCLSplitKernel(this);
+  if (use_preview_kernels) {
+    load_preview_kernels();
+  }
 }
 
 OpenCLDevice::~OpenCLDevice()
@@ -762,7 +771,7 @@ bool OpenCLDevice::load_kernels(const DeviceRequestedFeatures &requested_feature
   load_required_kernels(requested_features);
 
   vector<OpenCLProgram *> programs;
-  kernel_programs.load_kernels(programs, requested_features);
+  kernel_programs.load_kernels(programs, requested_features, false);
 
   if (!requested_features.use_baking && requested_features.use_denoising) {
     denoising_program = OpenCLProgram(
@@ -840,6 +849,19 @@ void OpenCLDevice::load_required_kernels(const DeviceRequestedFeatures &requeste
   }
 }
 
+void OpenCLDevice::load_preview_kernels()
+{
+  DeviceRequestedFeatures no_features;
+  vector<OpenCLProgram *> programs;
+  preview_programs.load_kernels(programs, no_features, true);
+
+  foreach (OpenCLProgram *program, programs) {
+    if (!program->load()) {
+      load_required_kernel_task_pool.push(function_bind(&OpenCLProgram::compile, program));
+    }
+  }
+}
+
 bool OpenCLDevice::wait_for_availability(const DeviceRequestedFeatures &requested_features)
 {
   if (requested_features.use_baking) {
@@ -847,18 +869,59 @@ bool OpenCLDevice::wait_for_availability(const DeviceRequestedFeatures &requeste
     return true;
   }
 
-  load_kernel_task_pool.wait_work();
+  if (background) {
+    load_kernel_task_pool.wait_work();
+    use_preview_kernels = false;
+  }
+  else {
+    /* We use a device setting to determine to load preview kernels or not
+     * Better to check on device level than per kernel as mixing preview and
+     * non-preview kernels does not work due to different data types */
+    if (use_preview_kernels) {
+      use_preview_kernels = load_kernel_num_compiling.load() > 0;
+    }
+  }
   return split_kernel->load_kernels(requested_features);
 }
 
 OpenCLDevice::OpenCLSplitPrograms *OpenCLDevice::get_split_programs()
 {
-  return &kernel_programs;
+  return use_preview_kernels ? &preview_programs : &kernel_programs;
 }
 
 DeviceKernelStatus OpenCLDevice::get_active_kernel_switch_state()
 {
-  return DEVICE_KERNEL_USING_FEATURE_KERNEL;
+  /* Do not switch kernels for background renderings
+   * We do foreground rendering but use the preview kernels
+   * Check for the optimized kernels
+   *
+   * This works also the other way around, where we are using
+   * optimized kernels but new ones are being compiled due
+   * to other features that are needed */
+  if (background) {
+    /* The if-statements below would find the same result,
+     * But as the `finished` method uses a mutex we added
+     * this as an early exit */
+    return DEVICE_KERNEL_USING_FEATURE_KERNEL;
+  }
+
+  bool other_kernels_finished = load_kernel_num_compiling.load() == 0;
+  if (use_preview_kernels) {
+    if (other_kernels_finished) {
+      return DEVICE_KERNEL_FEATURE_KERNEL_AVAILABLE;
+    }
+    else {
+      return DEVICE_KERNEL_WAITING_FOR_FEATURE_KERNEL;
+    }
+  }
+  else {
+    if (other_kernels_finished) {
+      return DEVICE_KERNEL_USING_FEATURE_KERNEL;
+    }
+    else {
+      return DEVICE_KERNEL_FEATURE_KERNEL_INVALID;
+    }
+  }
 }
 
 void OpenCLDevice::mem_alloc(device_memory &mem)

intern/cycles/graph/node.cpp

@@ -367,17 +367,9 @@ void Node::copy_value(const SocketType &socket, const Node &other, const SocketT
       case SocketType::TRANSFORM_ARRAY:
         copy_array<Transform>(this, socket, &other, other_socket);
         break;
-      case SocketType::NODE_ARRAY: {
+      case SocketType::NODE_ARRAY:
         copy_array<void *>(this, socket, &other, other_socket);
-
-        array<Node *> &node_array = get_socket_value<array<Node *>>(this, socket);
-
-        for (Node *node : node_array) {
-          node->reference();
-        }
-
         break;
-      }
       default:
         assert(0);
         break;
@@ -387,14 +379,6 @@ void Node::copy_value(const SocketType &socket, const Node &other, const SocketT
     const void *src = ((char *)&other) + other_socket.struct_offset;
     void *dst = ((char *)this) + socket.struct_offset;
     memcpy(dst, src, socket.size());
-
-    if (socket.type == SocketType::NODE) {
-      Node *node = get_socket_value<Node *>(this, socket);
-
-      if (node) {
-        node->reference();
-      }
-    }
   }
 }
 
@@ -789,26 +773,6 @@ void Node::set_owner(const NodeOwner *owner_)
   owner = owner_;
 }
 
-void Node::dereference_all_used_nodes()
-{
-  foreach (const SocketType &socket, type->inputs) {
-    if (socket.type == SocketType::NODE) {
-      Node *node = get_socket_value<Node *>(this, socket);
-
-      if (node) {
-        node->dereference();
-      }
-    }
-    else if (socket.type == SocketType::NODE_ARRAY) {
-      const array<Node *> &nodes = get_socket_value<array<Node *>>(this, socket);
-
-      for (Node *node : nodes) {
-        node->dereference();
-      }
-    }
-  }
-}
-
 bool Node::socket_is_modified(const SocketType &input) const
 {
   return (socket_modified & input.modified_flag_bit) != 0;
@@ -839,25 +803,6 @@ template<typename T> void Node::set_if_different(const SocketType &input, T valu
   socket_modified |= input.modified_flag_bit;
 }
 
-void Node::set_if_different(const SocketType &input, Node *value)
-{
-  if (get_socket_value<Node *>(this, input) == value) {
-    return;
-  }
-
-  Node *old_node = get_socket_value<Node *>(this, input);
-  if (old_node) {
-    old_node->dereference();
-  }
-
-  if (value) {
-    value->reference();
-  }
-
-  get_socket_value<Node *>(this, input) = value;
-  socket_modified |= input.modified_flag_bit;
-}
-
 template<typename T> void Node::set_if_different(const SocketType &input, array<T> &value)
 {
   if (!socket_is_modified(input)) {
@@ -870,27 +815,6 @@ template<typename T> void Node::set_if_different(const SocketType &input, array<
   socket_modified |= input.modified_flag_bit;
 }
 
-void Node::set_if_different(const SocketType &input, array<Node *> &value)
-{
-  if (!socket_is_modified(input)) {
-    if (get_socket_value<array<Node *>>(this, input) == value) {
-      return;
-    }
-  }
-
-  array<Node *> &old_nodes = get_socket_value<array<Node *>>(this, input);
-  for (Node *old_node : old_nodes) {
-    old_node->dereference();
-  }
-
-  for (Node *new_node : value) {
-    new_node->reference();
-  }
-
-  get_socket_value<array<Node *>>(this, input).steal_data(value);
-  socket_modified |= input.modified_flag_bit;
-}
-
 void Node::print_modified_sockets() const
 {
   printf("Node : %s\n", name.c_str());

intern/cycles/graph/node.h

@@ -177,32 +177,8 @@ struct Node {
   const NodeOwner *get_owner() const;
   void set_owner(const NodeOwner *owner_);
 
-  int reference_count() const
-  {
-    return ref_count;
-  }
-
-  void reference()
-  {
-    ref_count += 1;
-  }
-
-  void dereference()
-  {
-    ref_count -= 1;
-  }
-
-  /* Set the reference count to zero. This should only be called when we know for sure that the
-   * Node is not used by anyone else. For now, this is only the case when "deleting" shaders, as
-   * they are never actually deleted. */
-  void clear_reference_count()
-  {
-    ref_count = 0;
-  }
-
  protected:
   const NodeOwner *owner;
-  int ref_count{0};
 
   template<typename T> static T &get_socket_value(const Node *node, const SocketType &socket)
   {
@@ -213,19 +189,7 @@ struct Node {
 
   template<typename T> void set_if_different(const SocketType &input, T value);
 
-  /* Explicit overload for Node sockets so we can handle reference counting. The old Node is
-   * dereferenced, and the new one is referenced. */
-  void set_if_different(const SocketType &input, Node *value);
-
   template<typename T> void set_if_different(const SocketType &input, array<T> &value);
-
-  /* Explicit overload for Node sockets so we can handle reference counting. The old Nodes are
-   * dereferenced, and the new ones are referenced. */
-  void set_if_different(const SocketType &input, array<Node *> &value);
-
-  /* Call this function in derived classes' destructors to ensure that used Nodes are dereferenced
-   * properly. */
-  void dereference_all_used_nodes();
 };
 
 CCL_NAMESPACE_END

intern/cycles/graph/node_type.h

@@ -148,17 +148,16 @@ struct NodeType {
 #define NODE_DECLARE \
   static const NodeType *get_node_type(); \
   template<typename T> static const NodeType *register_type(); \
-  static Node *create(const NodeType *type); \
-  static const NodeType *node_type;
+  static Node *create(const NodeType *type);
 
 #define NODE_DEFINE(structname) \
-  const NodeType *structname::node_type = structname::register_type<structname>(); \
   Node *structname::create(const NodeType *) \
   { \
     return new structname(); \
   } \
   const NodeType *structname::get_node_type() \
   { \
+    static const NodeType *node_type = register_type<structname>(); \
     return node_type; \
   } \
   template<typename T> const NodeType *structname::register_type()
@@ -170,8 +169,6 @@ struct NodeType {
 #define NODE_ABSTRACT_DEFINE(structname) \
   const NodeType *structname::get_node_base_type() \
   { \
-    /* Base types constructed in this getter to ensure correct initialization \
-     * order. Regular types are not so they are auto-registered for XML parsing. */ \
     static const NodeType *node_base_type = register_base_type<structname>(); \
     return node_base_type; \
   } \

intern/cycles/kernel/closure/alloc.h

@@ -57,24 +57,14 @@ ccl_device ccl_addr_space void *closure_alloc_extra(ShaderData *sd, int size)
 
 ccl_device_inline ShaderClosure *bsdf_alloc(ShaderData *sd, int size, float3 weight)
 {
-  kernel_assert(isfinite3_safe(weight));
+  ShaderClosure *sc = closure_alloc(sd, size, CLOSURE_NONE_ID, weight);
 
-  const float sample_weight = fabsf(average(weight));
+  if (sc == NULL)
+    return NULL;
 
-  /* Use comparison this way to help dealing with non-finite weight: if the average is not finite
-   * we will not allocate new closure. */
-  if (sample_weight >= CLOSURE_WEIGHT_CUTOFF) {
-    ShaderClosure *sc = closure_alloc(sd, size, CLOSURE_NONE_ID, weight);
-    if (sc == NULL) {
-      return NULL;
-    }
-
-    sc->sample_weight = sample_weight;
-
-    return sc;
-  }
-
-  return NULL;
+  float sample_weight = fabsf(average(weight));
+  sc->sample_weight = sample_weight;
+  return (sample_weight >= CLOSURE_WEIGHT_CUTOFF) ? sc : NULL;
 }
 
 #ifdef __OSL__
@@ -83,27 +73,17 @@ ccl_device_inline ShaderClosure *bsdf_alloc_osl(ShaderData *sd,
                                                 float3 weight,
                                                 void *data)
 {
-  kernel_assert(isfinite3_safe(weight));
+  ShaderClosure *sc = closure_alloc(sd, size, CLOSURE_NONE_ID, weight);
 
-  const float sample_weight = fabsf(average(weight));
+  if (!sc)
+    return NULL;
 
-  /* Use comparison this way to help dealing with non-finite weight: if the average is not finite
-   * we will not allocate new closure. */
-  if (sample_weight >= CLOSURE_WEIGHT_CUTOFF) {
-    ShaderClosure *sc = closure_alloc(sd, size, CLOSURE_NONE_ID, weight);
-    if (!sc) {
-      return NULL;
-    }
+  memcpy((void *)sc, data, size);
 
-    memcpy((void *)sc, data, size);
-
-    sc->weight = weight;
-    sc->sample_weight = sample_weight;
-
-    return sc;
-  }
-
-  return NULL;
+  float sample_weight = fabsf(average(weight));
+  sc->weight = weight;
+  sc->sample_weight = sample_weight;
+  return (sample_weight >= CLOSURE_WEIGHT_CUTOFF) ? sc : NULL;
 }
 #endif
 

intern/cycles/kernel/kernel_light_common.h

@@ -200,12 +200,12 @@ ccl_device bool light_spread_clamp_area_light(const float3 P,
    * uv coordinates. */
   const float new_center_u = 0.5f * (min_u + max_u);
   const float new_center_v = 0.5f * (min_v + max_v);
-  const float new_len_u = max_u - min_u;
-  const float new_len_v = max_v - min_v;
+  const float new_len_u = 0.5f * (max_u - min_u);
+  const float new_len_v = 0.5f * (max_v - min_v);
 
   *lightP = *lightP + new_center_u * u + new_center_v * v;
-  *axisu = u * new_len_u;
-  *axisv = v * new_len_v;
+  *axisu = u * new_len_u * 2.0f;
+  *axisv = v * new_len_v * 2.0f;
 
   return true;
 }

intern/cycles/kernel/kernel_montecarlo.h

@@ -195,108 +195,31 @@ ccl_device float2 regular_polygon_sample(float corners, float rotation, float u,
 
 ccl_device float3 ensure_valid_reflection(float3 Ng, float3 I, float3 N)
 {
-  float3 R = 2 * dot(N, I) * N - I;
+  float3 R;
+  float NI = dot(N, I);
+  float NgR, threshold;
 
-  /* Reflection rays may always be at least as shallow as the incoming ray. */
-  float threshold = min(0.9f * dot(Ng, I), 0.01f);
-  if (dot(Ng, R) >= threshold) {
-    return N;
-  }
+  /* Check if the incident ray is coming from behind normal N. */
+  if (NI > 0) {
+    /* Normal reflection */
+    R = (2 * NI) * N - I;
+    NgR = dot(Ng, R);
 
-  /* Form coordinate system with Ng as the Z axis and N inside the X-Z-plane.
-   * The X axis is found by normalizing the component of N that's orthogonal to Ng.
-   * The Y axis isn't actually needed.
-   */
-  float NdotNg = dot(N, Ng);
-  float3 X = normalize(N - NdotNg * Ng);
-
-  /* Keep math expressions. */
-  /* clang-format off */
-  /* Calculate N.z and N.x in the local coordinate system.
-   *
-   * The goal of this computation is to find a N' that is rotated towards Ng just enough
-   * to lift R' above the threshold (here called t), therefore dot(R', Ng) = t.
-   *
-   * According to the standard reflection equation,
-   * this means that we want dot(2*dot(N', I)*N' - I, Ng) = t.
-   *
-   * Since the Z axis of our local coordinate system is Ng, dot(x, Ng) is just x.z, so we get
-   * 2*dot(N', I)*N'.z - I.z = t.
-   *
-   * The rotation is simple to express in the coordinate system we formed -
-   * since N lies in the X-Z-plane, we know that N' will also lie in the X-Z-plane,
-   * so N'.y = 0 and therefore dot(N', I) = N'.x*I.x + N'.z*I.z .
-   *
-   * Furthermore, we want N' to be normalized, so N'.x = sqrt(1 - N'.z^2).
-   *
-   * With these simplifications,
-   * we get the final equation 2*(sqrt(1 - N'.z^2)*I.x + N'.z*I.z)*N'.z - I.z = t.
-   *
-   * The only unknown here is N'.z, so we can solve for that.
-   *
-   * The equation has four solutions in general:
-   *
-   * N'.z = +-sqrt(0.5*(+-sqrt(I.x^2*(I.x^2 + I.z^2 - t^2)) + t*I.z + I.x^2 + I.z^2)/(I.x^2 + I.z^2))
-   * We can simplify this expression a bit by grouping terms:
-   *
-   * a = I.x^2 + I.z^2
-   * b = sqrt(I.x^2 * (a - t^2))
-   * c = I.z*t + a
-   * N'.z = +-sqrt(0.5*(+-b + c)/a)
-   *
-   * Two solutions can immediately be discarded because they're negative so N' would lie in the
-   * lower hemisphere.
-   */
-  /* clang-format on */
-
-  float Ix = dot(I, X), Iz = dot(I, Ng);
-  float Ix2 = sqr(Ix), Iz2 = sqr(Iz);
-  float a = Ix2 + Iz2;
-
-  float b = safe_sqrtf(Ix2 * (a - sqr(threshold)));
-  float c = Iz * threshold + a;
-
-  /* Evaluate both solutions.
-   * In many cases one can be immediately discarded (if N'.z would be imaginary or larger than
-   * one), so check for that first. If no option is viable (might happen in extreme cases like N
-   * being in the wrong hemisphere), give up and return Ng. */
-  float fac = 0.5f / a;
-  float N1_z2 = fac * (b + c), N2_z2 = fac * (-b + c);
-  bool valid1 = (N1_z2 > 1e-5f) && (N1_z2 <= (1.0f + 1e-5f));
-  bool valid2 = (N2_z2 > 1e-5f) && (N2_z2 <= (1.0f + 1e-5f));
-
-  float2 N_new;
-  if (valid1 && valid2) {
-    /* If both are possible, do the expensive reflection-based check. */
-    float2 N1 = make_float2(safe_sqrtf(1.0f - N1_z2), safe_sqrtf(N1_z2));
-    float2 N2 = make_float2(safe_sqrtf(1.0f - N2_z2), safe_sqrtf(N2_z2));
-
-    float R1 = 2 * (N1.x * Ix + N1.y * Iz) * N1.y - Iz;
-    float R2 = 2 * (N2.x * Ix + N2.y * Iz) * N2.y - Iz;
-
-    valid1 = (R1 >= 1e-5f);
-    valid2 = (R2 >= 1e-5f);
-    if (valid1 && valid2) {
-      /* If both solutions are valid, return the one with the shallower reflection since it will be
-       * closer to the input (if the original reflection wasn't shallow, we would not be in this
-       * part of the function). */
-      N_new = (R1 < R2) ? N1 : N2;
+    /* Reflection rays may always be at least as shallow as the incoming ray. */
+    threshold = min(0.9f * dot(Ng, I), 0.01f);
+    if (NgR >= threshold) {
+      return N;
     }
-    else {
-      /* If only one reflection is valid (= positive), pick that one. */
-      N_new = (R1 > R2) ? N1 : N2;
-    }
-  }
-  else if (valid1 || valid2) {
-    /* Only one solution passes the N'.z criterium, so pick that one. */
-    float Nz2 = valid1 ? N1_z2 : N2_z2;
-    N_new = make_float2(safe_sqrtf(1.0f - Nz2), safe_sqrtf(Nz2));
   }
   else {
-    return Ng;
+    /* Bad incident */
+    R = -I;
+    NgR = dot(Ng, R);
+    threshold = 0.01f;
   }
 
-  return N_new.x * X + N_new.y * Ng;
+  R = R + Ng * (threshold - NgR);            /* Lift the reflection above the threshold. */
+  return normalize(I * len(R) + R * len(I)); /* Find a bisector. */
 }
 
 CCL_NAMESPACE_END

intern/cycles/kernel/kernel_subsurface.h

@@ -25,9 +25,8 @@ CCL_NAMESPACE_BEGIN
 ccl_device_inline float3
 subsurface_scatter_eval(ShaderData *sd, const ShaderClosure *sc, float disk_r, float r, bool all)
 {
-  /* This is the Veach one-sample model with balance heuristic, some pdf
-   * factors drop out when using balance heuristic weighting. For branched
-   * path tracing (all) we sample all closure and don't use MIS. */
+  /* this is the veach one-sample model with balance heuristic, some pdf
+   * factors drop out when using balance heuristic weighting */
   float3 eval_sum = zero_float3();
   float pdf_sum = 0.0f;
   float sample_weight_inv = 0.0f;
@@ -66,30 +65,6 @@ subsurface_scatter_eval(ShaderData *sd, const ShaderClosure *sc, float disk_r, f
   return (pdf_sum > 0.0f) ? eval_sum / pdf_sum : zero_float3();
 }
 
-ccl_device_inline float3 subsurface_scatter_walk_eval(ShaderData *sd,
-                                                      const ShaderClosure *sc,
-                                                      float3 throughput,
-                                                      bool all)
-{
-  /* This is the Veach one-sample model with balance heuristic, some pdf
-   * factors drop out when using balance heuristic weighting. For branched
-   * path tracing (all) we sample all closure and don't use MIS. */
-  if (!all) {
-    float bssrdf_weight = 0.0f;
-    float weight = sc->sample_weight;
-
-    for (int i = 0; i < sd->num_closure; i++) {
-      sc = &sd->closure[i];
-
-      if (CLOSURE_IS_BSSRDF(sc->type)) {
-        bssrdf_weight += sc->sample_weight;
-      }
-    }
-    throughput *= bssrdf_weight / weight;
-  }
-  return throughput;
-}
-
 /* replace closures with a single diffuse bsdf closure after scatter step */
 ccl_device void subsurface_scatter_setup_diffuse_bsdf(
     KernelGlobals *kg, ShaderData *sd, ClosureType type, float roughness, float3 weight, float3 N)
@@ -462,8 +437,7 @@ ccl_device_noinline
                            ccl_addr_space PathState *state,
                            const ShaderClosure *sc,
                            const float bssrdf_u,
-                           const float bssrdf_v,
-                           bool all)
+                           const float bssrdf_v)
 {
   /* Sample diffuse surface scatter into the object. */
   float3 D;
@@ -632,10 +606,10 @@ ccl_device_noinline
       t = ray->t;
     }
     else if (bounce == 0) {
-      /* Restore original position if nothing was hit after the first bounce,
-       * without the ray_offset() that was added to avoid self-intersection.
-       * Otherwise if that offset is relatively large compared to the scattering
-       * radius, we never go back up high enough to exit the surface. */
+      /* Restore original position if nothing was hit after the first bounce.
+       * Otherwise if the ray_offset() to avoid self-intersection is relatively
+       * large compared to the scattering radius, we go never backup high enough
+       * to exit the surface. */
       ray->P = sd->P;
     }
 
@@ -695,7 +669,7 @@ ccl_device_noinline
   /* TODO: gain back performance lost from merging with disk BSSRDF. We
    * only need to return on hit so this indirect ray push/pop overhead
    * is not actually needed, but it does keep the code simpler. */
-  ss_isect->weight[0] = subsurface_scatter_walk_eval(sd, sc, throughput, all);
+  ss_isect->weight[0] = throughput;
 #ifdef __SPLIT_KERNEL__
   ss_isect->ray = *ray;
 #endif
@@ -717,7 +691,7 @@ ccl_device_inline int subsurface_scatter_multi_intersect(KernelGlobals *kg,
     return subsurface_scatter_disk(kg, ss_isect, sd, sc, lcg_state, bssrdf_u, bssrdf_v, all);
   }
   else {
-    return subsurface_random_walk(kg, ss_isect, sd, state, sc, bssrdf_u, bssrdf_v, all);
+    return subsurface_random_walk(kg, ss_isect, sd, state, sc, bssrdf_u, bssrdf_v);
   }
 }
 

intern/cycles/kernel/kernel_types.h

@@ -99,23 +99,27 @@ CCL_NAMESPACE_BEGIN
 #define __AO__
 #define __PASSES__
 #define __HAIR__
-#define __SVM__
-#define __EMISSION__
-#define __HOLDOUT__
-#define __MULTI_CLOSURE__
-#define __TRANSPARENT_SHADOWS__
-#define __BACKGROUND_MIS__
-#define __LAMP_MIS__
-#define __CAMERA_MOTION__
-#define __OBJECT_MOTION__
-#define __BAKING__
-#define __PRINCIPLED__
-#define __SUBSURFACE__
-#define __VOLUME__
-#define __VOLUME_SCATTER__
-#define __CMJ__
-#define __SHADOW_RECORD_ALL__
-#define __BRANCHED_PATH__
+
+/* Without these we get an AO render, used by OpenCL preview kernel. */
+#ifndef __KERNEL_AO_PREVIEW__
+#  define __SVM__
+#  define __EMISSION__
+#  define __HOLDOUT__
+#  define __MULTI_CLOSURE__
+#  define __TRANSPARENT_SHADOWS__
+#  define __BACKGROUND_MIS__
+#  define __LAMP_MIS__
+#  define __CAMERA_MOTION__
+#  define __OBJECT_MOTION__
+#  define __BAKING__
+#  define __PRINCIPLED__
+#  define __SUBSURFACE__
+#  define __VOLUME__
+#  define __VOLUME_SCATTER__
+#  define __CMJ__
+#  define __SHADOW_RECORD_ALL__
+#  define __BRANCHED_PATH__
+#endif
 
 /* Device specific features */
 #ifdef __KERNEL_CPU__

intern/cycles/kernel/shaders/node_noise_texture.osl

@@ -25,7 +25,7 @@
  * coordinates to act as a seed since the noise functions don't have seed values.
  * A seed value is needed for generating distortion textures and color outputs.
  * The offset's components are in the range [100, 200], not too high to cause
- * bad precision and not too small to be noticeable. We use float seed because
+ * bad precision and not to small to be noticeable. We use float seed because
  * OSL only support float hashes.
  */
 

intern/cycles/kernel/shaders/stdcycles.h

@@ -84,67 +84,30 @@ closure color principled_hair(normal N,
 closure color henyey_greenstein(float g) BUILTIN;
 closure color absorption() BUILTIN;
 
-normal ensure_valid_reflection(normal Ng, vector I, normal N)
+normal ensure_valid_reflection(normal Ng, normal I, normal N)
 {
   /* The implementation here mirrors the one in kernel_montecarlo.h,
    * check there for an explanation of the algorithm. */
+  vector R;
+  float NI = dot(N, I);
+  float NgR, threshold;
 
-  float sqr(float x)
-  {
-    return x * x;
-  }
-
-  vector R = 2 * dot(N, I) * N - I;
-
-  float threshold = min(0.9 * dot(Ng, I), 0.01);
-  if (dot(Ng, R) >= threshold) {
-    return N;
-  }
-
-  float NdotNg = dot(N, Ng);
-  vector X = normalize(N - NdotNg * Ng);
-
-  float Ix = dot(I, X), Iz = dot(I, Ng);
-  float Ix2 = sqr(Ix), Iz2 = sqr(Iz);
-  float a = Ix2 + Iz2;
-
-  float b = sqrt(Ix2 * (a - sqr(threshold)));
-  float c = Iz * threshold + a;
-
-  float fac = 0.5 / a;
-  float N1_z2 = fac * (b + c), N2_z2 = fac * (-b + c);
-  int valid1 = (N1_z2 > 1e-5) && (N1_z2 <= (1.0 + 1e-5));
-  int valid2 = (N2_z2 > 1e-5) && (N2_z2 <= (1.0 + 1e-5));
-
-  float N_new_x, N_new_z;
-  if (valid1 && valid2) {
-    float N1_x = sqrt(1.0 - N1_z2), N1_z = sqrt(N1_z2);
-    float N2_x = sqrt(1.0 - N2_z2), N2_z = sqrt(N2_z2);
-
-    float R1 = 2 * (N1_x * Ix + N1_z * Iz) * N1_z - Iz;
-    float R2 = 2 * (N2_x * Ix + N2_z * Iz) * N2_z - Iz;
-
-    valid1 = (R1 >= 1e-5);
-    valid2 = (R2 >= 1e-5);
-    if (valid1 && valid2) {
-      N_new_x = (R1 < R2) ? N1_x : N2_x;
-      N_new_z = (R1 < R2) ? N1_z : N2_z;
+  if (NI > 0) {
+    R = (2 * NI) * N - I;
+    NgR = dot(Ng, R);
+    threshold = min(0.9 * dot(Ng, I), 0.01);
+    if (NgR >= threshold) {
+      return N;
     }
-    else {
-      N_new_x = (R1 > R2) ? N1_x : N2_x;
-      N_new_z = (R1 > R2) ? N1_z : N2_z;
-    }
-  }
-  else if (valid1 || valid2) {
-    float Nz2 = valid1 ? N1_z2 : N2_z2;
-    N_new_x = sqrt(1.0 - Nz2);
-    N_new_z = sqrt(Nz2);
   }
   else {
-    return Ng;
+    R = -I;
+    NgR = dot(Ng, R);
+    threshold = 0.01;
   }
 
-  return N_new_x * X + N_new_z * Ng;
+  R = R + Ng * (threshold - NgR);
+  return normalize(I * length(R) + R * length(I));
 }
 
 #endif /* CCL_STDOSL_H */

intern/cycles/kernel/svm/svm_noisetex.h

@@ -20,7 +20,7 @@ CCL_NAMESPACE_BEGIN
  * coordinates to act as a seed since the noise functions don't have seed values.
  * A seed value is needed for generating distortion textures and color outputs.
  * The offset's components are in the range [100, 200], not too high to cause
- * bad precision and not too small to be noticeable. We use float seed because
+ * bad precision and not to small to be noticeable. We use float seed because
  * OSL only support float hashes.
  */
 

intern/cycles/kernel/svm/svm_tex_coord.h

@@ -370,13 +370,10 @@ ccl_device void svm_node_tangent(KernelGlobals *kg, ShaderData *sd, float *stack
 
   if (direction_type == NODE_TANGENT_UVMAP) {
     /* UV map */
-    if (desc.offset == ATTR_STD_NOT_FOUND) {
-      stack_store_float3(stack, tangent_offset, zero_float3());
-      return;
-    }
-    else {
+    if (desc.offset == ATTR_STD_NOT_FOUND)
+      tangent = make_float3(0.0f, 0.0f, 0.0f);
+    else
       tangent = attribute_value;
-    }
   }
   else {
     /* radial */

intern/cycles/render/CMakeLists.txt

@@ -24,7 +24,6 @@ set(INC_SYS
 
 set(SRC
   alembic.cpp
-  alembic_read.cpp
   attribute.cpp
   background.cpp
   bake.cpp
@@ -68,7 +67,6 @@ set(SRC
 
 set(SRC_HEADERS
   alembic.h
-  alembic_read.h
   attribute.h
   bake.h
   background.h

intern/cycles/render/alembic.h

@@ -152,10 +152,6 @@ template<typename T> class DataStore {
   double last_loaded_time = std::numeric_limits<double>::max();
 
  public:
-  /* Keys used to compare values. */
-  Alembic::AbcCoreAbstract::ArraySample::Key key1;
-  Alembic::AbcCoreAbstract::ArraySample::Key key2;
-
   void set_time_sampling(Alembic::AbcCoreAbstract::TimeSampling time_sampling_)
   {
     time_sampling = time_sampling_;
@@ -229,11 +225,6 @@ template<typename T> class DataStore {
     index_data_map.push_back({time, data_index.source_time, data_index.index});
   }
 
-  void add_no_data(double time)
-  {
-    index_data_map.push_back({time, time, -1ul});
-  }
-
   bool is_constant() const
   {
     return data.size() <= 1;
@@ -293,7 +284,7 @@ struct CachedData {
   DataStore<array<int3>> triangles{};
   /* triangle "loops" are the polygons' vertices indices used for indexing face varying attributes
    * (like UVs) */
-  DataStore<array<int>> uv_loops{};
+  DataStore<array<int3>> triangles_loops{};
   DataStore<array<int>> shader{};
 
   /* subd data */
@@ -371,18 +362,16 @@ class AlembicObject : public Node {
   void set_object(Object *object);
   Object *get_object();
 
-  void load_data_in_cache(CachedData &cached_data,
-                          AlembicProcedural *proc,
-                          Alembic::AbcGeom::IPolyMeshSchema &schema,
-                          Progress &progress);
-  void load_data_in_cache(CachedData &cached_data,
-                          AlembicProcedural *proc,
-                          Alembic::AbcGeom::ISubDSchema &schema,
-                          Progress &progress);
-  void load_data_in_cache(CachedData &cached_data,
-                          AlembicProcedural *proc,
-                          const Alembic::AbcGeom::ICurvesSchema &schema,
-                          Progress &progress);
+  void load_all_data(AlembicProcedural *proc,
+                     Alembic::AbcGeom::IPolyMeshSchema &schema,
+                     Progress &progress);
+  void load_all_data(AlembicProcedural *proc,
+                     Alembic::AbcGeom::ISubDSchema &schema,
+                     Progress &progress);
+  void load_all_data(AlembicProcedural *proc,
+                     const Alembic::AbcGeom::ICurvesSchema &schema,
+                     Progress &progress,
+                     float default_radius);
 
   bool has_data_loaded() const;
 
@@ -408,21 +397,33 @@ class AlembicObject : public Node {
 
   CachedData &get_cached_data()
   {
-    return cached_data_;
+    return cached_data;
   }
 
   bool is_constant() const
   {
-    return cached_data_.is_constant();
+    return cached_data.is_constant();
   }
 
   Object *object = nullptr;
 
   bool data_loaded = false;
 
-  CachedData cached_data_;
+  CachedData cached_data;
 
-  void setup_transform_cache(CachedData &cached_data, float scale);
+  void update_shader_attributes(const Alembic::AbcGeom::ICompoundProperty &arb_geom_params,
+                                Progress &progress);
+
+  void read_attribute(const Alembic::AbcGeom::ICompoundProperty &arb_geom_params,
+                      const ustring &attr_name,
+                      Progress &progress);
+
+  template<typename SchemaType>
+  void read_face_sets(SchemaType &schema,
+                      array<int> &polygon_to_shader,
+                      Alembic::AbcGeom::ISampleSelector sample_sel);
+
+  void setup_transform_cache(float scale);
 
   AttributeRequestSet get_requested_attributes();
 };

intern/cycles/render/alembic_read.h

@@ -1,134 +0,0 @@
-/*
- * Copyright 2021 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.
- */
-
-#pragma once
-
-#ifdef WITH_ALEMBIC
-
-#  include <Alembic/AbcCoreFactory/All.h>
-#  include <Alembic/AbcGeom/All.h>
-
-#  include "util/util_vector.h"
-
-CCL_NAMESPACE_BEGIN
-
-class AlembicProcedural;
-class AttributeRequestSet;
-class Progress;
-struct CachedData;
-
-/* Maps a FaceSet whose name matches that of a Shader to the index of said shader in the Geometry's
- * used_shaders list. */
-struct FaceSetShaderIndexPair {
-  Alembic::AbcGeom::IFaceSet face_set;
-  int shader_index;
-};
-
-/* Data of an IPolyMeshSchema that we need to read. */
-struct PolyMeshSchemaData {
-  Alembic::AbcGeom::TimeSamplingPtr time_sampling;
-  size_t num_samples;
-  Alembic::AbcGeom::MeshTopologyVariance topology_variance;
-
-  Alembic::AbcGeom::IP3fArrayProperty positions;
-  Alembic::AbcGeom::IInt32ArrayProperty face_indices;
-  Alembic::AbcGeom::IInt32ArrayProperty face_counts;
-
-  Alembic::AbcGeom::IN3fGeomParam normals;
-
-  vector<FaceSetShaderIndexPair> shader_face_sets;
-
-  // Unsupported for now.
-  Alembic::AbcGeom::IV3fArrayProperty velocities;
-};
-
-void read_geometry_data(AlembicProcedural *proc,
-                        CachedData &cached_data,
-                        const PolyMeshSchemaData &data,
-                        Progress &progress);
-
-/* Data of an ISubDSchema that we need to read. */
-struct SubDSchemaData {
-  Alembic::AbcGeom::TimeSamplingPtr time_sampling;
-  size_t num_samples;
-  Alembic::AbcGeom::MeshTopologyVariance topology_variance;
-
-  Alembic::AbcGeom::IInt32ArrayProperty face_counts;
-  Alembic::AbcGeom::IInt32ArrayProperty face_indices;
-  Alembic::AbcGeom::IP3fArrayProperty positions;
-
-  Alembic::AbcGeom::IInt32ArrayProperty crease_indices;
-  Alembic::AbcGeom::IInt32ArrayProperty crease_lengths;
-  Alembic::AbcGeom::IFloatArrayProperty crease_sharpnesses;
-
-  vector<FaceSetShaderIndexPair> shader_face_sets;
-
-  // Those are unsupported for now.
-  Alembic::AbcGeom::IInt32ArrayProperty corner_indices;
-  Alembic::AbcGeom::IFloatArrayProperty corner_sharpnesses;
-  Alembic::AbcGeom::IInt32Property face_varying_interpolate_boundary;
-  Alembic::AbcGeom::IInt32Property face_varying_propagate_corners;
-  Alembic::AbcGeom::IInt32Property interpolate_boundary;
-  Alembic::AbcGeom::IInt32ArrayProperty holes;
-  Alembic::AbcGeom::IStringProperty subdivision_scheme;
-  Alembic::AbcGeom::IV3fArrayProperty velocities;
-};
-
-void read_geometry_data(AlembicProcedural *proc,
-                        CachedData &cached_data,
-                        const SubDSchemaData &data,
-                        Progress &progress);
-
-/* Data of a ICurvesSchema that we need to read. */
-struct CurvesSchemaData {
-  Alembic::AbcGeom::TimeSamplingPtr time_sampling;
-  size_t num_samples;
-  Alembic::AbcGeom::MeshTopologyVariance topology_variance;
-
-  Alembic::AbcGeom::IP3fArrayProperty positions;
-
-  Alembic::AbcGeom::IInt32ArrayProperty num_vertices;
-
-  float default_radius;
-  float radius_scale;
-
-  // Those are unsupported for now.
-  Alembic::AbcGeom::IV3fArrayProperty velocities;
-  // if this property is invalid then the weight for every point is 1
-  Alembic::AbcGeom::IFloatArrayProperty position_weights;
-  Alembic::AbcGeom::IN3fGeomParam normals;
-  Alembic::AbcGeom::IFloatGeomParam widths;
-  Alembic::AbcGeom::IUcharArrayProperty orders;
-  Alembic::AbcGeom::IFloatArrayProperty knots;
-
-  // TODO(@kevindietrich): type, basis, wrap
-};
-
-void read_geometry_data(AlembicProcedural *proc,
-                        CachedData &cached_data,
-                        const CurvesSchemaData &data,
-                        Progress &progress);
-
-void read_attributes(AlembicProcedural *proc,
-                     CachedData &cache,
-                     const Alembic::AbcGeom::ICompoundProperty &arb_geom_params,
-                     const Alembic::AbcGeom::IV2fGeomParam &default_uvs_param,
-                     const AttributeRequestSet &requested_attributes,
-                     Progress &progress);
-
-CCL_NAMESPACE_END
-
-#endif

intern/cycles/render/attribute.cpp

@@ -383,23 +383,6 @@ AttributeStandard Attribute::name_standard(const char *name)
   return ATTR_STD_NONE;
 }
 
-AttrKernelDataType Attribute::kernel_type(const Attribute &attr)
-{
-  if (attr.element == ATTR_ELEMENT_CORNER) {
-    return AttrKernelDataType::UCHAR4;
-  }
-
-  if (attr.type == TypeDesc::TypeFloat) {
-    return AttrKernelDataType::FLOAT;
-  }
-
-  if (attr.type == TypeFloat2) {
-    return AttrKernelDataType::FLOAT2;
-  }
-
-  return AttrKernelDataType::FLOAT3;
-}
-
 void Attribute::get_uv_tiles(Geometry *geom,
                              AttributePrimitive prim,
                              unordered_set<int> &tiles) const
@@ -434,7 +417,7 @@ void Attribute::get_uv_tiles(Geometry *geom,
 /* Attribute Set */
 
 AttributeSet::AttributeSet(Geometry *geometry, AttributePrimitive prim)
-    : modified_flag(~0u), geometry(geometry), prim(prim)
+    : geometry(geometry), prim(prim)
 {
 }
 
@@ -457,7 +440,7 @@ Attribute *AttributeSet::add(ustring name, TypeDesc type, AttributeElement eleme
 
   Attribute new_attr(name, type, element, geometry, prim);
   attributes.emplace_back(std::move(new_attr));
-  tag_modified(attributes.back());
+  modified = true;
   return &attributes.back();
 }
 
@@ -479,7 +462,8 @@ void AttributeSet::remove(ustring name)
 
     for (it = attributes.begin(); it != attributes.end(); it++) {
       if (&*it == attr) {
-        remove(it);
+        modified = true;
+        attributes.erase(it);
         return;
       }
     }
@@ -624,7 +608,8 @@ void AttributeSet::remove(AttributeStandard std)
 
     for (it = attributes.begin(); it != attributes.end(); it++) {
       if (&*it == attr) {
-        remove(it);
+        modified = true;
+        attributes.erase(it);
         return;
       }
     }
@@ -649,12 +634,6 @@ void AttributeSet::remove(Attribute *attribute)
   }
 }
 
-void AttributeSet::remove(list<Attribute>::iterator it)
-{
-  tag_modified(*it);
-  attributes.erase(it);
-}
-
 void AttributeSet::resize(bool reserve_only)
 {
   foreach (Attribute &attr, attributes) {
@@ -695,22 +674,21 @@ void AttributeSet::update(AttributeSet &&new_attributes)
   for (it = attributes.begin(); it != attributes.end();) {
     if (it->std != ATTR_STD_NONE) {
       if (new_attributes.find(it->std) == nullptr) {
-        remove(it++);
+        modified = true;
+        attributes.erase(it++);
         continue;
       }
     }
     else if (it->name != "") {
       if (new_attributes.find(it->name) == nullptr) {
-        remove(it++);
+        modified = true;
+        attributes.erase(it++);
         continue;
       }
     }
 
     it++;
   }
-
-  /* If all attributes were replaced, transform is no longer applied. */
-  geometry->transform_applied = false;
 }
 
 void AttributeSet::clear_modified()
@@ -718,27 +696,7 @@ void AttributeSet::clear_modified()
   foreach (Attribute &attr, attributes) {
     attr.modified = false;
   }
-
-  modified_flag = 0;
-}
-
-void AttributeSet::tag_modified(const Attribute &attr)
-{
-  /* Some attributes are not stored in the various kernel attribute arrays
-   * (DeviceScene::attribute_*), so the modified flags are only set if the associated standard
-   * corresponds to an attribute which will be stored in the kernel's attribute arrays. */
-  const bool modifies_device_array = (attr.std != ATTR_STD_FACE_NORMAL &&
-                                      attr.std != ATTR_STD_VERTEX_NORMAL);
-
-  if (modifies_device_array) {
-    AttrKernelDataType kernel_type = Attribute::kernel_type(attr);
-    modified_flag |= (1u << kernel_type);
-  }
-}
-
-bool AttributeSet::modified(AttrKernelDataType kernel_type) const
-{
-  return (modified_flag & (1u << kernel_type)) != 0;
+  modified = false;
 }
 
 /* AttributeRequest */

intern/cycles/render/attribute.h

@@ -39,21 +39,6 @@ class Hair;
 class Mesh;
 struct Transform;
 
-/* AttrKernelDataType.
- *
- * The data type of the device arrays storing the attribute's data. Those data types are different
- * than the ones for attributes as some attribute types are stored in the same array, e.g. Point,
- * Vector, and Transform are all stored as float3 in the kernel.
- *
- * The values of this enumeration are also used as flags to detect changes in AttributeSet. */
-
-enum AttrKernelDataType {
-  FLOAT = 0,
-  FLOAT2 = 1,
-  FLOAT3 = 2,
-  UCHAR4 = 3,
-};
-
 /* Attribute
  *
  * Arbitrary data layers on meshes.
@@ -182,8 +167,6 @@ class Attribute {
   static const char *standard_name(AttributeStandard std);
   static AttributeStandard name_standard(const char *name);
 
-  static AttrKernelDataType kernel_type(const Attribute &attr);
-
   void get_uv_tiles(Geometry *geom, AttributePrimitive prim, unordered_set<int> &tiles) const;
 };
 
@@ -192,12 +175,11 @@ class Attribute {
  * Set of attributes on a mesh. */
 
 class AttributeSet {
-  uint32_t modified_flag;
-
  public:
   Geometry *geometry;
   AttributePrimitive prim;
   list<Attribute> attributes;
+  bool modified = true;
 
   AttributeSet(Geometry *geometry, AttributePrimitive prim);
   AttributeSet(AttributeSet &&) = default;
@@ -215,8 +197,6 @@ class AttributeSet {
 
   void remove(Attribute *attribute);
 
-  void remove(list<Attribute>::iterator it);
-
   void resize(bool reserve_only = false);
   void clear(bool preserve_voxel_data = false);
 
@@ -224,18 +204,7 @@ class AttributeSet {
    * and remove any attribute not found on the new set from this. */
   void update(AttributeSet &&new_attributes);
 
-  /* Return whether the attributes of the given kernel_type are modified, where "modified" means
-   * that some attributes of the given type were added or removed from this AttributeSet. This does
-   * not mean that the data of the remaining attributes in this AttributeSet were also modified. To
-   * check this, use Attribute.modified. */
-  bool modified(AttrKernelDataType kernel_type) const;
-
   void clear_modified();
-
- private:
-  /* Set the relevant modified flag for the attribute. Only attributes that are stored in device
-   * arrays will be considered for tagging this AttributeSet as modified. */
-  void tag_modified(const Attribute &attr);
 };
 
 /* AttributeRequest

intern/cycles/render/background.cpp

@@ -59,7 +59,6 @@ Background::Background() : Node(get_node_type())
 
 Background::~Background()
 {
-  dereference_all_used_nodes();
 }
 
 void Background::device_update(Device *device, DeviceScene *dscene, Scene *scene)

intern/cycles/render/geometry.cpp

@@ -46,12 +46,6 @@ CCL_NAMESPACE_BEGIN
 
 /* Geometry */
 
-PackFlags operator|=(PackFlags &pack_flags, uint32_t value)
-{
-  pack_flags = (PackFlags)((uint32_t)pack_flags | value);
-  return pack_flags;
-}
-
 NODE_ABSTRACT_DEFINE(Geometry)
 {
   NodeType *type = NodeType::add("geometry_base", NULL);
@@ -85,7 +79,6 @@ Geometry::Geometry(const NodeType *node_type, const Type type)
 
 Geometry::~Geometry()
 {
-  dereference_all_used_nodes();
   delete bvh;
 }
 
@@ -830,13 +823,10 @@ void GeometryManager::device_update_attributes(Device *device,
   dscene->attributes_float3.alloc(attr_float3_size);
   dscene->attributes_uchar4.alloc(attr_uchar4_size);
 
-  /* The order of those flags needs to match that of AttrKernelDataType. */
-  const bool attributes_need_realloc[4] = {
-      dscene->attributes_float.need_realloc(),
-      dscene->attributes_float2.need_realloc(),
-      dscene->attributes_float3.need_realloc(),
-      dscene->attributes_uchar4.need_realloc(),
-  };
+  const bool copy_all_data = dscene->attributes_float.need_realloc() ||
+                             dscene->attributes_float2.need_realloc() ||
+                             dscene->attributes_float3.need_realloc() ||
+                             dscene->attributes_uchar4.need_realloc();
 
   size_t attr_float_offset = 0;
   size_t attr_float2_offset = 0;
@@ -855,7 +845,7 @@ void GeometryManager::device_update_attributes(Device *device,
 
       if (attr) {
         /* force a copy if we need to reallocate all the data */
-        attr->modified |= attributes_need_realloc[Attribute::kernel_type(*attr)];
+        attr->modified |= copy_all_data;
       }
 
       update_attribute_element_offset(geom,
@@ -878,7 +868,7 @@ void GeometryManager::device_update_attributes(Device *device,
 
         if (subd_attr) {
           /* force a copy if we need to reallocate all the data */
-          subd_attr->modified |= attributes_need_realloc[Attribute::kernel_type(*subd_attr)];
+          subd_attr->modified |= copy_all_data;
         }
 
         update_attribute_element_offset(mesh,
@@ -909,10 +899,6 @@ void GeometryManager::device_update_attributes(Device *device,
     foreach (AttributeRequest &req, attributes.requests) {
       Attribute *attr = values.find(req);
 
-      if (attr) {
-        attr->modified |= attributes_need_realloc[Attribute::kernel_type(*attr)];
-      }
-
       update_attribute_element_offset(object->geometry,
                                       dscene->attributes_float,
                                       attr_float_offset,
@@ -948,10 +934,10 @@ void GeometryManager::device_update_attributes(Device *device,
   /* copy to device */
   progress.set_status("Updating Mesh", "Copying Attributes to device");
 
-  dscene->attributes_float.copy_to_device_if_modified();
-  dscene->attributes_float2.copy_to_device_if_modified();
-  dscene->attributes_float3.copy_to_device_if_modified();
-  dscene->attributes_uchar4.copy_to_device_if_modified();
+  dscene->attributes_float.copy_to_device();
+  dscene->attributes_float2.copy_to_device();
+  dscene->attributes_float3.copy_to_device();
+  dscene->attributes_uchar4.copy_to_device();
 
   if (progress.get_cancel())
     return;
@@ -1249,16 +1235,7 @@ void GeometryManager::device_update_bvh(Device *device,
 
   const bool can_refit = scene->bvh != nullptr &&
                          (bparams.bvh_layout == BVHLayout::BVH_LAYOUT_OPTIX);
-
-  PackFlags pack_flags = PackFlags::PACK_NONE;
-
-  if (scene->bvh == nullptr) {
-    pack_flags |= PackFlags::PACK_ALL;
-  }
-
-  if (dscene->prim_visibility.is_modified()) {
-    pack_flags |= PackFlags::PACK_VISIBILITY;
-  }
+  const bool pack_all = scene->bvh == nullptr;
 
   BVH *bvh = scene->bvh;
   if (!scene->bvh) {
@@ -1296,14 +1273,10 @@ void GeometryManager::device_update_bvh(Device *device,
 
     pack.root_index = -1;
 
-    if (pack_flags != PackFlags::PACK_ALL) {
+    if (!pack_all) {
       /* if we do not need to recreate the BVH, then only the vertices are updated, so we can
        * safely retake the memory */
       dscene->prim_tri_verts.give_data(pack.prim_tri_verts);
-
-      if ((pack_flags & PackFlags::PACK_VISIBILITY) != 0) {
-        dscene->prim_visibility.give_data(pack.prim_visibility);
-      }
     }
     else {
       /* It is not strictly necessary to skip those resizes we if do not have to repack, as the OS
@@ -1332,21 +1305,13 @@ void GeometryManager::device_update_bvh(Device *device,
     // Iterate over scene mesh list instead of objects, since 'optix_prim_offset' was calculated
     // based on that list, which may be ordered differently from the object list.
     foreach (Geometry *geom, scene->geometry) {
-      /* Make a copy of the pack_flags so the current geometry's flags do not pollute the others'.
-       */
-      PackFlags geom_pack_flags = pack_flags;
-
-      if (geom->is_modified()) {
-        geom_pack_flags |= PackFlags::PACK_VERTICES;
-      }
-
-      if (geom_pack_flags == PACK_NONE) {
+      if (!pack_all && !geom->is_modified()) {
         continue;
       }
 
       const pair<int, uint> &info = geometry_to_object_info[geom];
       pool.push(function_bind(
-          &Geometry::pack_primitives, geom, &pack, info.first, info.second, geom_pack_flags));
+          &Geometry::pack_primitives, geom, &pack, info.first, info.second, pack_all));
     }
     pool.wait_work();
   }
@@ -1381,7 +1346,7 @@ void GeometryManager::device_update_bvh(Device *device,
     dscene->prim_type.steal_data(pack.prim_type);
     dscene->prim_type.copy_to_device();
   }
-  if (pack.prim_visibility.size() && (dscene->prim_visibility.is_modified() || has_bvh2_layout)) {
+  if (pack.prim_visibility.size() && (dscene->prim_visibility.need_realloc() || has_bvh2_layout)) {
     dscene->prim_visibility.steal_data(pack.prim_visibility);
     dscene->prim_visibility.copy_to_device();
   }
@@ -1438,43 +1403,21 @@ static void update_device_flags_attribute(uint32_t &device_update_flags,
       continue;
     }
 
-    AttrKernelDataType kernel_type = Attribute::kernel_type(attr);
-
-    switch (kernel_type) {
-      case AttrKernelDataType::FLOAT: {
-        device_update_flags |= ATTR_FLOAT_MODIFIED;
-        break;
-      }
-      case AttrKernelDataType::FLOAT2: {
-        device_update_flags |= ATTR_FLOAT2_MODIFIED;
-        break;
-      }
-      case AttrKernelDataType::FLOAT3: {
-        device_update_flags |= ATTR_FLOAT3_MODIFIED;
-        break;
-      }
-      case AttrKernelDataType::UCHAR4: {
-        device_update_flags |= ATTR_UCHAR4_MODIFIED;
-        break;
-      }
+    if (attr.element == ATTR_ELEMENT_CORNER) {
+      device_update_flags |= ATTR_UCHAR4_MODIFIED;
+    }
+    else if (attr.type == TypeDesc::TypeFloat) {
+      device_update_flags |= ATTR_FLOAT_MODIFIED;
+    }
+    else if (attr.type == TypeFloat2) {
+      device_update_flags |= ATTR_FLOAT2_MODIFIED;
+    }
+    else if (attr.type == TypeDesc::TypeMatrix) {
+      device_update_flags |= ATTR_FLOAT3_MODIFIED;
+    }
+    else if (attr.element != ATTR_ELEMENT_VOXEL) {
+      device_update_flags |= ATTR_FLOAT3_MODIFIED;
     }
-  }
-}
-
-static void update_attribute_realloc_flags(uint32_t &device_update_flags,
-                                           const AttributeSet &attributes)
-{
-  if (attributes.modified(AttrKernelDataType::FLOAT)) {
-    device_update_flags |= ATTR_FLOAT_NEEDS_REALLOC;
-  }
-  if (attributes.modified(AttrKernelDataType::FLOAT2)) {
-    device_update_flags |= ATTR_FLOAT2_NEEDS_REALLOC;
-  }
-  if (attributes.modified(AttrKernelDataType::FLOAT3)) {
-    device_update_flags |= ATTR_FLOAT3_NEEDS_REALLOC;
-  }
-  if (attributes.modified(AttrKernelDataType::UCHAR4)) {
-    device_update_flags |= ATTR_UCHAR4_NEEDS_REALLOC;
   }
 }
 
@@ -1500,11 +1443,16 @@ void GeometryManager::device_update_preprocess(Device *device, Scene *scene, Pro
   foreach (Geometry *geom, scene->geometry) {
     geom->has_volume = false;
 
-    update_attribute_realloc_flags(device_update_flags, geom->attributes);
+    if (geom->attributes.modified) {
+      device_update_flags |= ATTRS_NEED_REALLOC;
+    }
 
     if (geom->is_mesh()) {
       Mesh *mesh = static_cast<Mesh *>(geom);
-      update_attribute_realloc_flags(device_update_flags, mesh->subd_attributes);
+
+      if (mesh->subd_attributes.modified) {
+        device_update_flags |= ATTRS_NEED_REALLOC;
+      }
     }
 
     foreach (Node *node, geom->get_used_shaders()) {
@@ -1646,10 +1594,6 @@ void GeometryManager::device_update_preprocess(Device *device, Scene *scene, Pro
     }
   }
 
-  if ((update_flags & VISIBILITY_MODIFIED) != 0) {
-    dscene->prim_visibility.tag_modified();
-  }
-
   if (device_update_flags & ATTR_FLOAT_NEEDS_REALLOC) {
     dscene->attributes_map.tag_realloc();
     dscene->attributes_float.tag_realloc();
@@ -1976,8 +1920,7 @@ void GeometryManager::device_update(Device *device,
    * Also update the BVH if the transformations change, we cannot rely on tagging the Geometry
    * as modified in this case, as we may accumulate displacement if the vertices do not also
    * change. */
-  bool need_update_scene_bvh = (scene->bvh == nullptr ||
-                                (update_flags & (TRANSFORM_MODIFIED | VISIBILITY_MODIFIED)) != 0);
+  bool need_update_scene_bvh = (scene->bvh == nullptr || (update_flags & TRANSFORM_MODIFIED) != 0);
   {
     scoped_callback_timer timer([scene](double time) {
       if (scene->update_stats) {
@@ -2063,7 +2006,7 @@ void GeometryManager::device_update(Device *device,
      * for meshes with correct bounding boxes.
      *
      * This wouldn't cause wrong results, just true
-     * displacement might be less optimal to calculate.
+     * displacement might be less optimal ot calculate.
      */
     scene->object_manager->need_flags_update = old_need_object_flags_update;
   }

intern/cycles/render/geometry.h

@@ -43,24 +43,6 @@ class Shader;
 class Volume;
 struct PackedBVH;
 
-/* Flags used to determine which geometry data need to be packed. */
-enum PackFlags : uint32_t {
-  PACK_NONE = 0u,
-
-  /* Pack the geometry information (e.g. triangle or curve keys indices). */
-  PACK_GEOMETRY = (1u << 0),
-
-  /* Pack the vertices, for Meshes and Volumes' bounding meshes. */
-  PACK_VERTICES = (1u << 1),
-
-  /* Pack the visibility flags for each triangle or curve. */
-  PACK_VISIBILITY = (1u << 2),
-
-  PACK_ALL = (PACK_GEOMETRY | PACK_VERTICES | PACK_VISIBILITY),
-};
-
-PackFlags operator|=(PackFlags &pack_flags, uint32_t value);
-
 /* Geometry
  *
  * Base class for geometric types like Mesh and Hair. */
@@ -144,10 +126,7 @@ class Geometry : public Node {
                    int n,
                    int total);
 
-  virtual void pack_primitives(PackedBVH *pack,
-                               int object,
-                               uint visibility,
-                               PackFlags pack_flags) = 0;
+  virtual void pack_primitives(PackedBVH *pack, int object, uint visibility, bool pack_all) = 0;
 
   /* Check whether the geometry should have own BVH built separately. Briefly,
    * own BVH is needed for geometry, if:
@@ -212,8 +191,6 @@ class GeometryManager {
 
     TRANSFORM_MODIFIED = (1 << 10),
 
-    VISIBILITY_MODIFIED = (1 << 11),
-
     /* tag everything in the manager for an update */
     UPDATE_ALL = ~0u,
 

intern/cycles/render/hair.cpp

@@ -494,47 +494,38 @@ void Hair::pack_curves(Scene *scene,
   }
 }
 
-void Hair::pack_primitives(PackedBVH *pack, int object, uint visibility, PackFlags pack_flags)
+void Hair::pack_primitives(PackedBVH *pack, int object, uint visibility, bool pack_all)
 {
   if (curve_first_key.empty())
     return;
 
-  /* Separate loop as other arrays are not initialized if their packing is not required. */
-  if ((pack_flags & PACK_VISIBILITY) != 0) {
-    unsigned int *prim_visibility = &pack->prim_visibility[optix_prim_offset];
-
-    size_t index = 0;
-    for (size_t j = 0; j < num_curves(); ++j) {
-      Curve curve = get_curve(j);
-      for (size_t k = 0; k < curve.num_segments(); ++k, ++index) {
-        prim_visibility[index] = visibility;
-      }
-    }
+  /* If the BVH does not have to be recreated, we can bail out. */
+  if (!pack_all) {
+    return;
   }
 
-  if ((pack_flags & PACK_GEOMETRY) != 0) {
-    unsigned int *prim_tri_index = &pack->prim_tri_index[optix_prim_offset];
-    int *prim_type = &pack->prim_type[optix_prim_offset];
-    int *prim_index = &pack->prim_index[optix_prim_offset];
-    int *prim_object = &pack->prim_object[optix_prim_offset];
-    // 'pack->prim_time' is unused by Embree and OptiX
+  unsigned int *prim_tri_index = &pack->prim_tri_index[optix_prim_offset];
+  int *prim_type = &pack->prim_type[optix_prim_offset];
+  unsigned int *prim_visibility = &pack->prim_visibility[optix_prim_offset];
+  int *prim_index = &pack->prim_index[optix_prim_offset];
+  int *prim_object = &pack->prim_object[optix_prim_offset];
+  // 'pack->prim_time' is unused by Embree and OptiX
 
-    uint type = has_motion_blur() ?
-                    ((curve_shape == CURVE_RIBBON) ? PRIMITIVE_MOTION_CURVE_RIBBON :
-                                                     PRIMITIVE_MOTION_CURVE_THICK) :
-                    ((curve_shape == CURVE_RIBBON) ? PRIMITIVE_CURVE_RIBBON :
-                                                     PRIMITIVE_CURVE_THICK);
+  uint type = has_motion_blur() ?
+                  ((curve_shape == CURVE_RIBBON) ? PRIMITIVE_MOTION_CURVE_RIBBON :
+                                                   PRIMITIVE_MOTION_CURVE_THICK) :
+                  ((curve_shape == CURVE_RIBBON) ? PRIMITIVE_CURVE_RIBBON : PRIMITIVE_CURVE_THICK);
 
-    size_t index = 0;
-    for (size_t j = 0; j < num_curves(); ++j) {
-      Curve curve = get_curve(j);
-      for (size_t k = 0; k < curve.num_segments(); ++k, ++index) {
-        prim_tri_index[index] = -1;
-        prim_type[index] = PRIMITIVE_PACK_SEGMENT(type, k);
-        // Each curve segment points back to its curve index
-        prim_index[index] = j + prim_offset;
-        prim_object[index] = object;
-      }
+  size_t index = 0;
+  for (size_t j = 0; j < num_curves(); ++j) {
+    Curve curve = get_curve(j);
+    for (size_t k = 0; k < curve.num_segments(); ++k, ++index) {
+      prim_tri_index[index] = -1;
+      prim_type[index] = PRIMITIVE_PACK_SEGMENT(type, k);
+      prim_visibility[index] = visibility;
+      // Each curve segment points back to its curve index
+      prim_index[index] = j + prim_offset;
+      prim_object[index] = object;
     }
   }
 }

intern/cycles/render/hair.h

@@ -146,10 +146,7 @@ class Hair : public Geometry {
   /* BVH */
   void pack_curves(Scene *scene, float4 *curve_key_co, float4 *curve_data, size_t curvekey_offset);
 
-  void pack_primitives(PackedBVH *pack,
-                       int object,
-                       uint visibility,
-                       PackFlags pack_flags) override;
+  void pack_primitives(PackedBVH *pack, int object, uint visibility, bool pack_all) override;
 };
 
 CCL_NAMESPACE_END

intern/cycles/render/light.cpp

@@ -159,7 +159,6 @@ NODE_DEFINE(Light)
 
 Light::Light() : Node(get_node_type())
 {
-  dereference_all_used_nodes();
 }
 
 void Light::tag_update(Scene *scene)
@@ -865,7 +864,7 @@ void LightManager::device_update_points(Device *, DeviceScene *dscene, Scene *sc
       const float min_spread_angle = 1.0f * M_PI_F / 180.0f;
       const float spread_angle = 0.5f * (M_PI_F - max(light->spread, min_spread_angle));
       /* Normalization computed using:
-       * integrate cos(x) * (1 - tan(x) * tan(a)) * sin(x) from x = 0 to pi/2 - a. */
+       * integrate cos(x) (1 - tan(x) * tan(a)) * sin(x) from x = a to pi/2. */
       const float tan_spread = tanf(spread_angle);
       const float normalize_spread = 2.0f / (2.0f + (2.0f * spread_angle - M_PI_F) * tan_spread);
 

intern/cycles/render/mesh.cpp

@@ -805,7 +805,7 @@ void Mesh::pack_patches(uint *patch_data, uint vert_offset, uint face_offset, ui
   }
 }
 
-void Mesh::pack_primitives(ccl::PackedBVH *pack, int object, uint visibility, PackFlags pack_flags)
+void Mesh::pack_primitives(ccl::PackedBVH *pack, int object, uint visibility, bool pack_all)
 {
   if (triangles.empty())
     return;
@@ -819,38 +819,28 @@ void Mesh::pack_primitives(ccl::PackedBVH *pack, int object, uint visibility, Pa
 
   uint type = has_motion_blur() ? PRIMITIVE_MOTION_TRIANGLE : PRIMITIVE_TRIANGLE;
 
-  /* Separate loop as other arrays are not initialized if their packing is not required. */
-  if ((pack_flags & PackFlags::PACK_VISIBILITY) != 0) {
-    unsigned int *prim_visibility = &pack->prim_visibility[optix_prim_offset];
-    for (size_t k = 0; k < num_prims; ++k) {
-      prim_visibility[k] = visibility;
-    }
-  }
-
-  if ((pack_flags & PackFlags::PACK_GEOMETRY) != 0) {
+  if (pack_all) {
     /* Use optix_prim_offset for indexing as those arrays also contain data for Hair geometries. */
     unsigned int *prim_tri_index = &pack->prim_tri_index[optix_prim_offset];
     int *prim_type = &pack->prim_type[optix_prim_offset];
+    unsigned int *prim_visibility = &pack->prim_visibility[optix_prim_offset];
     int *prim_index = &pack->prim_index[optix_prim_offset];
     int *prim_object = &pack->prim_object[optix_prim_offset];
 
     for (size_t k = 0; k < num_prims; ++k) {
-      if ((pack_flags & PackFlags::PACK_GEOMETRY) != 0) {
-        prim_tri_index[k] = (prim_offset + k) * 3;
-        prim_type[k] = type;
-        prim_index[k] = prim_offset + k;
-        prim_object[k] = object;
-      }
+      prim_tri_index[k] = (prim_offset + k) * 3;
+      prim_type[k] = type;
+      prim_index[k] = prim_offset + k;
+      prim_object[k] = object;
+      prim_visibility[k] = visibility;
     }
   }
 
-  if ((pack_flags & PackFlags::PACK_VERTICES) != 0) {
-    for (size_t k = 0; k < num_prims; ++k) {
-      const Mesh::Triangle t = get_triangle(k);
-      prim_tri_verts[k * 3] = float3_to_float4(verts[t.v[0]]);
-      prim_tri_verts[k * 3 + 1] = float3_to_float4(verts[t.v[1]]);
-      prim_tri_verts[k * 3 + 2] = float3_to_float4(verts[t.v[2]]);
-    }
+  for (size_t k = 0; k < num_prims; ++k) {
+    const Mesh::Triangle t = get_triangle(k);
+    prim_tri_verts[k * 3] = float3_to_float4(verts[t.v[0]]);
+    prim_tri_verts[k * 3 + 1] = float3_to_float4(verts[t.v[1]]);
+    prim_tri_verts[k * 3 + 2] = float3_to_float4(verts[t.v[2]]);
   }
 }
 

intern/cycles/render/mesh.h

@@ -232,10 +232,7 @@ class Mesh : public Geometry {
                   size_t tri_offset);
   void pack_patches(uint *patch_data, uint vert_offset, uint face_offset, uint corner_offset);
 
-  void pack_primitives(PackedBVH *pack,
-                       int object,
-                       uint visibility,
-                       PackFlags pack_flags) override;
+  void pack_primitives(PackedBVH *pack, int object, uint visibility, bool pack_all) override;
 
   void tessellate(DiagSplit *split);
 

intern/cycles/render/nodes.h

@@ -1600,23 +1600,11 @@ class SetNormalNode : public ShaderNode {
   NODE_SOCKET_API(float3, direction)
 };
 
-class OSLNode final : public ShaderNode {
+class OSLNode : public ShaderNode {
  public:
   static OSLNode *create(ShaderGraph *graph, size_t num_inputs, const OSLNode *from = NULL);
   ~OSLNode();
 
-  static void operator delete(void *ptr)
-  {
-    /* Override delete operator to silence new-delete-type-mismatch ASAN warnings
-     * regarding size mismatch in the destructor. This is intentional as we allocate
-     * extra space at the end of the node. */
-    ::operator delete(ptr);
-  }
-  static void operator delete(void *, void *)
-  {
-    /* Deliberately empty placement delete operator, to avoid MSVC warning C4291. */
-  }
-
   ShaderNode *clone(ShaderGraph *graph) const;
 
   char *input_default_value();

intern/cycles/render/object.cpp

@@ -220,10 +220,6 @@ void Object::tag_update(Scene *scene)
       flag |= ObjectManager::TRANSFORM_MODIFIED;
     }
 
-    if (visibility_is_modified()) {
-      flag |= ObjectManager::VISIBILITY_MODIFIED;
-    }
-
     foreach (Node *node, geometry->get_used_shaders()) {
       Shader *shader = static_cast<Shader *>(node);
       if (shader->get_use_mis() && shader->has_surface_emission)
@@ -918,10 +914,6 @@ void ObjectManager::tag_update(Scene *scene, uint32_t flag)
       geometry_flag |= GeometryManager::TRANSFORM_MODIFIED;
     }
 
-    if ((flag & VISIBILITY_MODIFIED) != 0) {
-      geometry_flag |= GeometryManager::VISIBILITY_MODIFIED;
-    }
-
     scene->geometry_manager->tag_update(scene, geometry_flag);
   }
 

intern/cycles/render/object.h

@@ -134,7 +134,6 @@ class ObjectManager {
     OBJECT_MODIFIED = (1 << 5),
     HOLDOUT_MODIFIED = (1 << 6),
     TRANSFORM_MODIFIED = (1 << 7),
-    VISIBILITY_MODIFIED = (1 << 8),
 
     /* tag everything in the manager for an update */
     UPDATE_ALL = ~0u,

intern/cycles/render/osl.cpp

@@ -91,10 +91,10 @@ void OSLShaderManager::reset(Scene * /*scene*/)
   shading_system_init();
 }
 
-void OSLShaderManager::device_update_specific(Device *device,
-                                              DeviceScene *dscene,
-                                              Scene *scene,
-                                              Progress &progress)
+void OSLShaderManager::device_update(Device *device,
+                                     DeviceScene *dscene,
+                                     Scene *scene,
+                                     Progress &progress)
 {
   if (!need_update())
     return;
@@ -1149,7 +1149,7 @@ void OSLCompiler::compile(OSLGlobals *og, Shader *shader)
     shader->has_integrator_dependency = false;
 
     /* generate surface shader */
-    if (shader->reference_count() && graph && output->input("Surface")->link) {
+    if (shader->used && graph && output->input("Surface")->link) {
       shader->osl_surface_ref = compile_type(shader, shader->graph, SHADER_TYPE_SURFACE);
 
       if (has_bump)
@@ -1165,7 +1165,7 @@ void OSLCompiler::compile(OSLGlobals *og, Shader *shader)
     }
 
     /* generate volume shader */
-    if (shader->reference_count() && graph && output->input("Volume")->link) {
+    if (shader->used && graph && output->input("Volume")->link) {
       shader->osl_volume_ref = compile_type(shader, shader->graph, SHADER_TYPE_VOLUME);
       shader->has_volume = true;
     }
@@ -1173,7 +1173,7 @@ void OSLCompiler::compile(OSLGlobals *og, Shader *shader)
       shader->osl_volume_ref = OSL::ShaderGroupRef();
 
     /* generate displacement shader */
-    if (shader->reference_count() && graph && output->input("Displacement")->link) {
+    if (shader->used && graph && output->input("Displacement")->link) {
       shader->osl_displacement_ref = compile_type(shader, shader->graph, SHADER_TYPE_DISPLACEMENT);
       shader->has_displacement = true;
     }

intern/cycles/render/osl.h

@@ -72,18 +72,15 @@ class OSLShaderManager : public ShaderManager {
 
   static void free_memory();
 
-  void reset(Scene *scene) override;
+  void reset(Scene *scene);
 
-  bool use_osl() override
+  bool use_osl()
   {
     return true;
   }
 
-  void device_update_specific(Device *device,
-                              DeviceScene *dscene,
-                              Scene *scene,
-                              Progress &progress) override;
-  void device_free(Device *device, DeviceScene *dscene, Scene *scene) override;
+  void device_update(Device *device, DeviceScene *dscene, Scene *scene, Progress &progress);
+  void device_free(Device *device, DeviceScene *dscene, Scene *scene);
 
   /* osl compile and query */
   static bool osl_compile(const string &inputfile, const string &outputfile);

intern/cycles/render/scene.cpp

@@ -143,27 +143,21 @@ void Scene::free_memory(bool final)
   delete bvh;
   bvh = NULL;
 
-  /* The order of deletion is important to make sure data is freed based on possible dependencies
-   * as the Nodes' reference counts are decremented in the destructors:
-   *
-   * - Procedurals can create and hold pointers to any other types.
-   * - Objects can hold pointers to Geometries and ParticleSystems
-   * - Lights and Geometries can hold pointers to Shaders.
-   *
-   * Similarly, we first delete all nodes and their associated device data, and then the managers
-   * and their associated device data.
-   */
+  foreach (Shader *s, shaders)
+    delete s;
+  /* delete procedurals before other types as they may hold pointers to those types */
   foreach (Procedural *p, procedurals)
     delete p;
-  foreach (Object *o, objects)
-    delete o;
   foreach (Geometry *g, geometry)
     delete g;
-  foreach (ParticleSystem *p, particle_systems)
-    delete p;
+  foreach (Object *o, objects)
+    delete o;
   foreach (Light *l, lights)
     delete l;
+  foreach (ParticleSystem *p, particle_systems)
+    delete p;
 
+  shaders.clear();
   geometry.clear();
   objects.clear();
   lights.clear();
@@ -175,25 +169,7 @@ void Scene::free_memory(bool final)
     film->device_free(device, &dscene, this);
     background->device_free(device, &dscene);
     integrator->device_free(device, &dscene, true);
-  }
 
-  if (final) {
-    delete camera;
-    delete dicing_camera;
-    delete film;
-    delete background;
-    delete integrator;
-  }
-
-  /* Delete Shaders after every other nodes to ensure that we do not try to decrement the reference
-   * count on some dangling pointer. */
-  foreach (Shader *s, shaders)
-    delete s;
-
-  shaders.clear();
-
-  /* Now that all nodes have been deleted, we can safely delete managers and device data. */
-  if (device) {
     object_manager->device_free(device, &dscene, true);
     geometry_manager->device_free(device, &dscene, true);
     shader_manager->device_free(device, &dscene, this);
@@ -213,6 +189,11 @@ void Scene::free_memory(bool final)
 
   if (final) {
     delete lookup_tables;
+    delete camera;
+    delete dicing_camera;
+    delete film;
+    delete background;
+    delete integrator;
     delete object_manager;
     delete geometry_manager;
     delete shader_manager;
@@ -523,6 +504,9 @@ bool Scene::update(Progress &progress, bool &kernel_switch_needed)
 {
   /* update scene */
   if (need_update()) {
+    /* Updated used shader tag so we know which features are need for the kernel. */
+    shader_manager->update_shaders_used(this);
+
     /* Update max_closures. */
     KernelIntegrator *kintegrator = &dscene.data.integrator;
     if (params.background) {
@@ -542,6 +526,9 @@ bool Scene::update(Progress &progress, bool &kernel_switch_needed)
     DeviceKernelStatus kernel_switch_status = device->get_active_kernel_switch_state();
     kernel_switch_needed = kernel_switch_status == DEVICE_KERNEL_FEATURE_KERNEL_AVAILABLE ||
                            kernel_switch_status == DEVICE_KERNEL_FEATURE_KERNEL_INVALID;
+    if (kernel_switch_status == DEVICE_KERNEL_WAITING_FOR_FEATURE_KERNEL) {
+      progress.set_kernel_status("Compiling render kernels");
+    }
     if (new_kernels_needed || kernel_switch_needed) {
       progress.set_kernel_status("Compiling render kernels");
       device->wait_for_availability(loaded_kernel_features);
@@ -579,6 +566,9 @@ bool Scene::load_kernels(Progress &progress, bool lock_scene)
       return false;
     }
 
+    progress.add_skip_time(timer, false);
+    VLOG(1) << "Total time spent loading kernels: " << time_dt() - timer.get_start();
+
     kernels_loaded = true;
     loaded_kernel_features = requested_features;
     return true;
@@ -597,7 +587,7 @@ int Scene::get_max_closure_count()
   int max_closures = 0;
   for (int i = 0; i < shaders.size(); i++) {
     Shader *shader = shaders[i];
-    if (shader->reference_count()) {
+    if (shader->used) {
       int num_closures = shader->graph->get_num_closures();
       max_closures = max(max_closures, num_closures);
     }
@@ -758,10 +748,9 @@ template<> void Scene::delete_node_impl(ParticleSystem *node)
   particle_system_manager->tag_update(this);
 }
 
-template<> void Scene::delete_node_impl(Shader *shader)
+template<> void Scene::delete_node_impl(Shader * /*node*/)
 {
   /* don't delete unused shaders, not supported */
-  shader->clear_reference_count();
 }
 
 template<> void Scene::delete_node_impl(Procedural *node)
@@ -828,12 +817,9 @@ template<> void Scene::delete_nodes(const set<ParticleSystem *> &nodes, const No
   particle_system_manager->tag_update(this);
 }
 
-template<> void Scene::delete_nodes(const set<Shader *> &nodes, const NodeOwner * /*owner*/)
+template<> void Scene::delete_nodes(const set<Shader *> & /*nodes*/, const NodeOwner * /*owner*/)
 {
   /* don't delete unused shaders, not supported */
-  for (Shader *shader : nodes) {
-    shader->clear_reference_count();
-  }
 }
 
 template<> void Scene::delete_nodes(const set<Procedural *> &nodes, const NodeOwner *owner)

intern/cycles/render/session.cpp

@@ -243,6 +243,11 @@ void Session::run_gpu()
       }
     }
 
+    /* Don't go in pause mode when image was rendered with preview kernels
+     * When feature kernels become available the session will be reset. */
+    else if (no_tiles && kernel_state == DEVICE_KERNEL_WAITING_FOR_FEATURE_KERNEL) {
+      time_sleep(0.1);
+    }
     else if (no_tiles && kernel_state == DEVICE_KERNEL_FEATURE_KERNEL_AVAILABLE) {
       reset_gpu(tile_manager.params, params.samples);
     }
@@ -757,6 +762,11 @@ void Session::run_cpu()
       }
     }
 
+    /* Don't go in pause mode when preview kernels are used
+     * When feature kernels become available the session will be reset. */
+    else if (no_tiles && kernel_state == DEVICE_KERNEL_WAITING_FOR_FEATURE_KERNEL) {
+      time_sleep(0.1);
+    }
     else if (no_tiles && kernel_state == DEVICE_KERNEL_FEATURE_KERNEL_AVAILABLE) {
       reset_cpu(tile_manager.params, params.samples);
     }

intern/cycles/render/shader.cpp

@@ -16,6 +16,7 @@
 
 #include "device/device.h"
 
+#include "render/alembic.h"
 #include "render/background.h"
 #include "render/camera.h"
 #include "render/colorspace.h"
@@ -26,7 +27,6 @@
 #include "render/nodes.h"
 #include "render/object.h"
 #include "render/osl.h"
-#include "render/procedural.h"
 #include "render/scene.h"
 #include "render/shader.h"
 #include "render/svm.h"
@@ -218,6 +218,7 @@ Shader::Shader() : Node(get_node_type())
   displacement_method = DISPLACE_BUMP;
 
   id = -1;
+  used = false;
 
   need_update_uvs = true;
   need_update_attribute = true;
@@ -381,9 +382,8 @@ void Shader::tag_used(Scene *scene)
 {
   /* if an unused shader suddenly gets used somewhere, it needs to be
    * recompiled because it was skipped for compilation before */
-  if (!reference_count()) {
+  if (!used) {
     tag_modified();
-    /* We do not reference here as the shader will be referenced when added to a socket. */
     scene->shader_manager->tag_update(scene, ShaderManager::SHADER_MODIFIED);
   }
 }
@@ -461,28 +461,52 @@ int ShaderManager::get_shader_id(Shader *shader, bool smooth)
   return id;
 }
 
-void ShaderManager::device_update(Device *device,
-                                  DeviceScene *dscene,
-                                  Scene *scene,
-                                  Progress &progress)
+void ShaderManager::update_shaders_used(Scene *scene)
 {
   if (!need_update()) {
     return;
   }
 
+  /* figure out which shaders are in use, so SVM/OSL can skip compiling them
+   * for speed and avoid loading image textures into memory */
   uint id = 0;
   foreach (Shader *shader, scene->shaders) {
+    shader->used = false;
     shader->id = id++;
   }
 
-  /* Those shaders should always be compiled as they are used as fallback if a shader cannot be
-   * found, e.g. bad shader index for the triangle shaders on a Mesh. */
-  assert(scene->default_surface->reference_count() != 0);
-  assert(scene->default_light->reference_count() != 0);
-  assert(scene->default_background->reference_count() != 0);
-  assert(scene->default_empty->reference_count() != 0);
+  scene->default_surface->used = true;
+  scene->default_light->used = true;
+  scene->default_background->used = true;
+  scene->default_empty->used = true;
 
-  device_update_specific(device, dscene, scene, progress);
+  if (scene->background->get_shader())
+    scene->background->get_shader()->used = true;
+
+#ifdef WITH_ALEMBIC
+  foreach (Procedural *procedural, scene->procedurals) {
+    AlembicProcedural *abc_proc = static_cast<AlembicProcedural *>(procedural);
+
+    foreach (Node *abc_node, abc_proc->get_objects()) {
+      AlembicObject *abc_object = static_cast<AlembicObject *>(abc_node);
+
+      foreach (Node *node, abc_object->get_used_shaders()) {
+        Shader *shader = static_cast<Shader *>(node);
+        shader->used = true;
+      }
+    }
+  }
+#endif
+
+  foreach (Geometry *geom, scene->geometry)
+    foreach (Node *node, geom->get_used_shaders()) {
+      Shader *shader = static_cast<Shader *>(node);
+      shader->used = true;
+    }
+
+  foreach (Light *light, scene->lights)
+    if (light->get_shader())
+      const_cast<Shader *>(light->get_shader())->used = true;
 }
 
 void ShaderManager::device_update_common(Device *device,
@@ -615,7 +639,6 @@ void ShaderManager::add_default(Scene *scene)
     Shader *shader = scene->create_node<Shader>();
     shader->name = "default_surface";
     shader->set_graph(graph);
-    shader->reference();
     scene->default_surface = shader;
     shader->tag_update(scene);
   }
@@ -634,8 +657,6 @@ void ShaderManager::add_default(Scene *scene)
     shader->set_graph(graph);
     scene->default_volume = shader;
     shader->tag_update(scene);
-    /* No default reference for the volume to avoid compiling volume kernels if there are no actual
-     * volumes in the scene */
   }
 
   /* default light */
@@ -652,7 +673,6 @@ void ShaderManager::add_default(Scene *scene)
     Shader *shader = scene->create_node<Shader>();
     shader->name = "default_light";
     shader->set_graph(graph);
-    shader->reference();
     scene->default_light = shader;
     shader->tag_update(scene);
   }
@@ -664,7 +684,6 @@ void ShaderManager::add_default(Scene *scene)
     Shader *shader = scene->create_node<Shader>();
     shader->name = "default_background";
     shader->set_graph(graph);
-    shader->reference();
     scene->default_background = shader;
     shader->tag_update(scene);
   }
@@ -676,7 +695,6 @@ void ShaderManager::add_default(Scene *scene)
     Shader *shader = scene->create_node<Shader>();
     shader->name = "default_empty";
     shader->set_graph(graph);
-    shader->reference();
     scene->default_empty = shader;
     shader->tag_update(scene);
   }
@@ -717,7 +735,7 @@ void ShaderManager::get_requested_features(Scene *scene,
   requested_features->nodes_features = 0;
   for (int i = 0; i < scene->shaders.size(); i++) {
     Shader *shader = scene->shaders[i];
-    if (!shader->reference_count()) {
+    if (!shader->used) {
       continue;
     }