Cleanup Pep8

This was duplicated

.clang-format

@@ -180,6 +180,7 @@ ForEachMacros:
   - CTX_DATA_BEGIN_WITH_ID
   - DEG_OBJECT_ITER_BEGIN
   - DEG_OBJECT_ITER_FOR_RENDER_ENGINE_BEGIN
+  - DRW_ENABLED_ENGINE_ITER
   - DRIVER_TARGETS_LOOPER_BEGIN
   - DRIVER_TARGETS_USED_LOOPER_BEGIN
   - FOREACH_BASE_IN_EDIT_MODE_BEGIN
@@ -263,6 +264,8 @@ ForEachMacros:
   - SET_SLOT_PROBING_BEGIN
   - MAP_SLOT_PROBING_BEGIN
   - VECTOR_SET_SLOT_PROBING_BEGIN
+  - ITEM_FOREACH_BEGIN
+  - ITEM_FOREACH_BEGIN_NO_CULL
 
 StatementMacros:
   - PyObject_HEAD

CMakeLists.txt

@@ -403,7 +403,7 @@ option(WITH_CYCLES_CUDA_BINARIES    "Build Cycles CUDA binaries" OFF)
 option(WITH_CYCLES_CUBIN_COMPILER   "Build cubins with nvrtc based compiler instead of nvcc" OFF)
 option(WITH_CYCLES_CUDA_BUILD_SERIAL "Build cubins one after another (useful on machines with limited RAM)" OFF)
 mark_as_advanced(WITH_CYCLES_CUDA_BUILD_SERIAL)
-set(CYCLES_TEST_DEVICES CPU CACHE STRING "Run regression tests on the specified device types (CPU CUDA OPTIX OPENCL)" )
+set(CYCLES_TEST_DEVICES CPU CACHE STRING "Run regression tests on the specified device types (CPU CUDA OPTIX)" )
 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 "CUDA architectures to build binaries for")
 mark_as_advanced(CYCLES_CUDA_BINARIES_ARCH)
 unset(PLATFORM_DEFAULT)
@@ -418,12 +418,10 @@ mark_as_advanced(WITH_CYCLES_DEBUG_NAN)
 mark_as_advanced(WITH_CYCLES_NATIVE_ONLY)
 
 option(WITH_CYCLES_DEVICE_CUDA              "Enable Cycles CUDA compute support" ON)
-option(WITH_CYCLES_DEVICE_OPTIX             "Enable Cycles OptiX support" OFF)
-option(WITH_CYCLES_DEVICE_OPENCL            "Enable Cycles OpenCL compute support" ON)
-option(WITH_CYCLES_NETWORK              "Enable Cycles compute over network support (EXPERIMENTAL and unfinished)" OFF)
+option(WITH_CYCLES_DEVICE_OPTIX             "Enable Cycles OptiX support" ON)
+option(WITH_CYCLES_DEVICE_HIP               "Enable Cycles HIP support" OFF)
+mark_as_advanced(WITH_CYCLES_DEVICE_HIP)
 mark_as_advanced(WITH_CYCLES_DEVICE_CUDA)
-mark_as_advanced(WITH_CYCLES_DEVICE_OPENCL)
-mark_as_advanced(WITH_CYCLES_NETWORK)
 
 option(WITH_CUDA_DYNLOAD "Dynamically load CUDA libraries at runtime" ON)
 mark_as_advanced(WITH_CUDA_DYNLOAD)
@@ -825,6 +823,11 @@ if(NOT WITH_CUDA_DYNLOAD)
   endif()
 endif()
 
+if(WITH_CYCLES_DEVICE_HIP)
+  # Currently HIP must be dynamically loaded, this may change in future toolkits
+  set(WITH_HIP_DYNLOAD ON)
+endif()
+
 #-----------------------------------------------------------------------------
 # Check check if submodules are cloned
 
@@ -1854,6 +1857,9 @@ elseif(WITH_CYCLES_STANDALONE)
   if(WITH_CUDA_DYNLOAD)
     add_subdirectory(extern/cuew)
   endif()
+  if(WITH_HIP_DYNLOAD)
+    add_subdirectory(extern/hipew)
+  endif()
   if(NOT WITH_SYSTEM_GLEW)
     add_subdirectory(extern/glew)
   endif()

build_files/build_environment/patches/ffmpeg.diff

@@ -70,16 +70,18 @@
  }
 --- a/libavcodec/rl.c
 +++ b/libavcodec/rl.c
-@@ -71,7 +71,7 @@ av_cold void ff_rl_init(RLTable *rl,
+@@ -71,17 +71,19 @@
  av_cold void ff_rl_init_vlc(RLTable *rl, unsigned static_size)
  {
      int i, q;
 -    VLC_TYPE table[1500][2] = {{0}};
 +    VLC_TYPE (*table)[2] = av_calloc(sizeof(VLC_TYPE), 1500 * 2);
      VLC vlc = { .table = table, .table_allocated = static_size };
-     av_assert0(static_size <= FF_ARRAY_ELEMS(table));
+-    av_assert0(static_size <= FF_ARRAY_ELEMS(table));
++    av_assert0(static_size < 1500);
      init_vlc(&vlc, 9, rl->n + 1, &rl->table_vlc[0][1], 4, 2, &rl->table_vlc[0][0], 4, 2, INIT_VLC_USE_NEW_STATIC);
-@@ -80,8 +80,10 @@ av_cold void ff_rl_init_vlc(RLTable *rl, unsigned static_size)
+ 
+     for (q = 0; q < 32; q++) {
          int qmul = q * 2;
          int qadd = (q - 1) | 1;
  
@@ -91,7 +93,7 @@
  
          if (q == 0) {
              qmul = 1;
-@@ -113,4 +115,5 @@ av_cold void ff_rl_init_vlc(RLTable *rl, unsigned static_size)
+@@ -113,4 +115,5 @@
              rl->rl_vlc[q][i].run   = run;
          }
      }

build_files/cmake/Modules/FindOptiX.cmake

@@ -33,11 +33,23 @@ FIND_PATH(OPTIX_INCLUDE_DIR
     include
 )
 
+IF(EXISTS "${OPTIX_INCLUDE_DIR}/optix.h")
+  FILE(STRINGS "${OPTIX_INCLUDE_DIR}/optix.h" _optix_version REGEX "^#define OPTIX_VERSION[ \t].*$")
+  STRING(REGEX MATCHALL "[0-9]+" _optix_version ${_optix_version})
+
+  MATH(EXPR _optix_version_major "${_optix_version} / 10000")
+  MATH(EXPR _optix_version_minor "(${_optix_version} % 10000) / 100")
+  MATH(EXPR _optix_version_patch "${_optix_version} % 100")
+
+  SET(OPTIX_VERSION "${_optix_version_major}.${_optix_version_minor}.${_optix_version_patch}")
+ENDIF()
+
 # handle the QUIETLY and REQUIRED arguments and set OPTIX_FOUND to TRUE if
 # all listed variables are TRUE
 INCLUDE(FindPackageHandleStandardArgs)
-FIND_PACKAGE_HANDLE_STANDARD_ARGS(OptiX DEFAULT_MSG
-    OPTIX_INCLUDE_DIR)
+FIND_PACKAGE_HANDLE_STANDARD_ARGS(OptiX
+    REQUIRED_VARS OPTIX_INCLUDE_DIR
+    VERSION_VAR OPTIX_VERSION)
 
 IF(OPTIX_FOUND)
   SET(OPTIX_INCLUDE_DIRS ${OPTIX_INCLUDE_DIR})
@@ -45,6 +57,7 @@ ENDIF()
 
 MARK_AS_ADVANCED(
   OPTIX_INCLUDE_DIR
+  OPTIX_VERSION
 )
 
 UNSET(_optix_SEARCH_DIRS)

build_files/config/pipeline_config.yaml

@@ -55,7 +55,7 @@ buildbot:
     cuda11:
         version: '11.4.1'
     optix:
-        version: '7.1.0'
+        version: '7.3.0'
     cmake:
         default:
             version: any

build_files/utils/make_update.py

@@ -200,15 +200,20 @@ def submodules_update(args, release_version, branch):
             if msg:
                 skip_msg += submodule_path + " skipped: " + msg + "\n"
             else:
-                if make_utils.git_branch(args.git_command) != submodule_branch:
-                    call([args.git_command, "fetch", "origin"])
-                    call([args.git_command, "checkout", submodule_branch])
-                call([args.git_command, "pull", "--rebase", "origin", submodule_branch])
-                # If we cannot find the specified branch for this submodule, fallback to default one (aka master).
-                if make_utils.git_branch(args.git_command) != submodule_branch:
-                    call([args.git_command, "fetch", "origin"])
-                    call([args.git_command, "checkout", submodule_branch_fallback])
-                    call([args.git_command, "pull", "--rebase", "origin", submodule_branch_fallback])
+                # Find a matching branch that exists.
+                call([args.git_command, "fetch", "origin"])
+                if make_utils.git_branch_exists(args.git_command, submodule_branch):
+                    pass
+                elif make_utils.git_branch_exists(args.git_command, submodule_branch_fallback):
+                    submodule_branch = submodule_branch_fallback
+                else:
+                    submodule_branch = None
+
+                # Switch to branch and pull.
+                if submodule_branch:
+                    if make_utils.git_branch(args.git_command) != submodule_branch:
+                        call([args.git_command, "checkout", submodule_branch])
+                    call([args.git_command, "pull", "--rebase", "origin", submodule_branch])
         finally:
             os.chdir(cwd)
 
@@ -222,6 +227,10 @@ if __name__ == "__main__":
 
     # Test if we are building a specific release version.
     branch = make_utils.git_branch(args.git_command)
+    if branch == 'HEAD':
+        sys.stderr.write('Blender git repository is in detached HEAD state, must be in a branch\n')
+        sys.exit(1)
+
     tag = make_utils.git_tag(args.git_command)
     release_version = make_utils.git_branch_release_version(branch, tag)
 

build_files/utils/make_utils.py

@@ -8,14 +8,19 @@ import subprocess
 import sys
 
 
-def call(cmd, exit_on_error=True):
-    print(" ".join(cmd))
+def call(cmd, exit_on_error=True, silent=False):
+    if not silent:
+        print(" ".join(cmd))
 
     # Flush to ensure correct order output on Windows.
     sys.stdout.flush()
     sys.stderr.flush()
 
-    retcode = subprocess.call(cmd)
+    if silent:
+        retcode = subprocess.call(cmd, stdout=subprocess.DEVNULL, stderr=subprocess.DEVNULL)
+    else:
+        retcode = subprocess.call(cmd)
+
     if exit_on_error and retcode != 0:
         sys.exit(retcode)
     return retcode
@@ -38,6 +43,11 @@ def check_output(cmd, exit_on_error=True):
     return output.strip()
 
 
+def git_branch_exists(git_command, branch):
+    return call([git_command, "rev-parse", "--verify", branch], exit_on_error=False, silent=True) == 0 or \
+           call([git_command, "rev-parse", "--verify", "remotes/origin/" + branch], exit_on_error=False, silent=True) == 0
+
+
 def git_branch(git_command):
     # Get current branch name.
     try:
@@ -70,7 +80,7 @@ def git_branch_release_version(branch, tag):
     return release_version
 
 
-def svn_libraries_base_url(release_version, branch):
+def svn_libraries_base_url(release_version, branch=None):
     if release_version:
         svn_branch = "tags/blender-" + release_version + "-release"
     elif branch:

doc/python_api/examples/aud.py

@@ -14,7 +14,7 @@ sound = aud.Sound('music.ogg')
 # play the audio, this return a handle to control play/pause
 handle = device.play(sound)
 # if the audio is not too big and will be used often you can buffer it
-sound_buffered = aud.Sound.buffer(sound)
+sound_buffered = aud.Sound.cache(sound)
 handle_buffered = device.play(sound_buffered)
 
 # stop the sounds (otherwise they play until their ends)

doc/python_api/requirements.txt

@@ -10,4 +10,4 @@ requests==2.26.0
 
 # Only needed to match the theme used for the official documentation.
 # Without this theme, the default theme will be used.
-sphinx_rtd_theme==1.0.0rc1
+sphinx_rtd_theme==1.0.0

extern/CMakeLists.txt

@@ -67,9 +67,12 @@ endif()
 
 if(WITH_CYCLES OR WITH_COMPOSITOR OR WITH_OPENSUBDIV)
   add_subdirectory(clew)
-  if(WITH_CUDA_DYNLOAD)
+  if((WITH_CYCLES_DEVICE_CUDA OR WITH_CYCLES_DEVICE_OPTIX) AND WITH_CUDA_DYNLOAD)
     add_subdirectory(cuew)
   endif()
+  if(WITH_CYCLES_DEVICE_HIP AND WITH_HIP_DYNLOAD)
+    add_subdirectory(hipew)
+  endif()
 endif()
 
 if(WITH_GHOST_X11 AND WITH_GHOST_XDND)

extern/audaspace/CMakeLists.txt

@@ -129,6 +129,7 @@ set(SRC
 	src/util/Barrier.cpp
 	src/util/Buffer.cpp
 	src/util/BufferReader.cpp
+	src/util/RingBuffer.cpp
 	src/util/StreamBuffer.cpp
 	src/util/ThreadPool.cpp
 )
@@ -152,6 +153,7 @@ set(PUBLIC_HDR
 	include/devices/ThreadedDevice.h
 	include/Exception.h
 	include/file/File.h
+	include/file/FileInfo.h
 	include/file/FileManager.h
 	include/file/FileWriter.h
 	include/file/IFileInput.h
@@ -244,6 +246,7 @@ set(PUBLIC_HDR
 	include/util/BufferReader.h
 	include/util/ILockable.h
 	include/util/Math3D.h
+	include/util/RingBuffer.h
 	include/util/StreamBuffer.h
 	include/util/ThreadPool.h
 )
@@ -960,7 +963,10 @@ endif()
 if(BUILD_DEMOS)
 	include_directories(${INCLUDE})
 
-	set(DEMOS audaplay audaconvert audaremap signalgen randsounds dynamicmusic playbackmanager)
+	set(DEMOS audainfo audaplay audaconvert audaremap signalgen randsounds dynamicmusic playbackmanager)
+
+	add_executable(audainfo demos/audainfo.cpp)
+	target_link_libraries(audainfo audaspace)
 
 	add_executable(audaplay demos/audaplay.cpp)
 	target_link_libraries(audaplay audaspace)

extern/audaspace/bindings/C/AUD_PlaybackManager.h

@@ -39,7 +39,7 @@ extern AUD_API void AUD_PlaybackManager_free(AUD_PlaybackManager* manager);
 * Plays a sound through the playback manager, adding it into a category.
 * \param manager The PlaybackManager object.
 * \param sound The sound to be played.
-* \param catKey The key of the category into which the sound will be added. If it doesn't exist a new one will be creatd.
+* \param catKey The key of the category into which the sound will be added. If it doesn't exist a new one will be created.
 */
 extern AUD_API void AUD_PlaybackManager_play(AUD_PlaybackManager* manager, AUD_Sound* sound, unsigned int catKey);
 

extern/audaspace/bindings/C/AUD_Sound.cpp

@@ -94,6 +94,40 @@ AUD_API int AUD_Sound_getLength(AUD_Sound* sound)
 	return (*sound)->createReader()->getLength();
 }
 
+AUD_API int AUD_Sound_getFileStreams(AUD_Sound* sound, AUD_StreamInfo **stream_infos)
+{
+	assert(sound);
+
+	std::shared_ptr<File> file = std::dynamic_pointer_cast<File>(*sound);
+
+	if(file)
+	{
+		try
+		{
+			auto streams = file->queryStreams();
+
+			size_t size = sizeof(AUD_StreamInfo) * streams.size();
+
+			if(!size)
+			{
+				*stream_infos = nullptr;
+				return 0;
+			}
+
+			*stream_infos = reinterpret_cast<AUD_StreamInfo*>(std::malloc(size));
+			std::memcpy(*stream_infos, streams.data(), size);
+
+			return streams.size();
+		}
+		catch(Exception&)
+		{
+		}
+	}
+
+	*stream_infos = nullptr;
+	return 0;
+}
+
 AUD_API sample_t* AUD_Sound_data(AUD_Sound* sound, int* length, AUD_Specs* specs)
 {
 	assert(sound);
@@ -252,6 +286,12 @@ AUD_API AUD_Sound* AUD_Sound_bufferFile(unsigned char* buffer, int size)
 	return new AUD_Sound(new File(buffer, size));
 }
 
+AUD_API AUD_Sound* AUD_Sound_bufferFileStream(unsigned char* buffer, int size, int stream)
+{
+	assert(buffer);
+	return new AUD_Sound(new File(buffer, size, stream));
+}
+
 AUD_API AUD_Sound* AUD_Sound_cache(AUD_Sound* sound)
 {
 	assert(sound);
@@ -272,6 +312,12 @@ AUD_API AUD_Sound* AUD_Sound_file(const char* filename)
 	return new AUD_Sound(new File(filename));
 }
 
+AUD_API AUD_Sound* AUD_Sound_fileStream(const char* filename, int stream)
+{
+	assert(filename);
+	return new AUD_Sound(new File(filename, stream));
+}
+
 AUD_API AUD_Sound* AUD_Sound_sawtooth(float frequency, AUD_SampleRate rate)
 {
 	return new AUD_Sound(new Sawtooth(frequency, rate));

extern/audaspace/bindings/C/AUD_Sound.h

@@ -36,7 +36,15 @@ extern AUD_API AUD_Specs AUD_Sound_getSpecs(AUD_Sound* sound);
  * \return The length of the sound in samples.
  * \note This function creates a reader from the sound and deletes it again.
  */
-extern AUD_API int AUD_getLength(AUD_Sound* sound);
+extern AUD_API int AUD_Sound_getLength(AUD_Sound* sound);
+
+/**
+ * Retrieves the stream infos of a sound file.
+ * \param sound The sound to retrieve from which must be a file sound.
+ * \param infos A pointer to a AUD_StreamInfo array that will be allocated and must afterwards be freed by the caller.
+ * \return The number of items in the infos array.
+ */
+extern AUD_API int AUD_Sound_getFileStreams(AUD_Sound* sound, AUD_StreamInfo** stream_infos);
 
 /**
  * Reads a sound's samples into memory.
@@ -89,6 +97,15 @@ extern AUD_API AUD_Sound* AUD_Sound_buffer(sample_t* data, int length, AUD_Specs
  */
 extern AUD_API AUD_Sound* AUD_Sound_bufferFile(unsigned char* buffer, int size);
 
+/**
+ * Loads a sound file from a memory buffer.
+ * \param buffer The buffer which contains the sound file.
+ * \param size The size of the buffer.
+ * \param stream The index of the audio stream within the file if it contains multiple audio streams.
+ * \return A handle of the sound file.
+ */
+extern AUD_API AUD_Sound* AUD_Sound_bufferFileStream(unsigned char* buffer, int size, int stream);
+
 /**
  * Caches a sound into a memory buffer.
  * \param sound The sound to cache.
@@ -103,6 +120,14 @@ extern AUD_API AUD_Sound* AUD_Sound_cache(AUD_Sound* sound);
  */
 extern AUD_API AUD_Sound* AUD_Sound_file(const char* filename);
 
+/**
+ * Loads a sound file.
+ * \param filename The filename of the sound file.
+ * \param stream The index of the audio stream within the file if it contains multiple audio streams.
+ * \return A handle of the sound file.
+ */
+extern AUD_API AUD_Sound* AUD_Sound_fileStream(const char* filename, int stream);
+
 /**
  * Creates a sawtooth sound.
  * \param frequency The frequency of the generated sawtooth sound.

extern/audaspace/bindings/C/AUD_Special.cpp

@@ -86,7 +86,6 @@ AUD_API AUD_SoundInfo AUD_getInfo(AUD_Sound* sound)
 	info.specs.channels = AUD_CHANNELS_INVALID;
 	info.specs.rate = AUD_RATE_INVALID;
 	info.length = 0.0f;
-	info.start_offset = 0.0f;
 
 	try
 	{
@@ -96,7 +95,6 @@ AUD_API AUD_SoundInfo AUD_getInfo(AUD_Sound* sound)
 		{
 			info.specs = convSpecToC(reader->getSpecs());
 			info.length = reader->getLength() / (float) info.specs.rate;
-			info.start_offset = reader->getStartOffset();
 		}
 	}
 	catch(Exception&)
@@ -109,7 +107,7 @@ AUD_API AUD_SoundInfo AUD_getInfo(AUD_Sound* sound)
 AUD_API float* AUD_readSoundBuffer(const char* filename, float low, float high,
 						   float attack, float release, float threshold,
 						   int accumulate, int additive, int square,
-						   float sthreshold, double samplerate, int* length)
+						   float sthreshold, double samplerate, int* length, int stream)
 {
 	Buffer buffer;
 	DeviceSpecs specs;
@@ -117,7 +115,7 @@ AUD_API float* AUD_readSoundBuffer(const char* filename, float low, float high,
 	specs.rate = (SampleRate)samplerate;
 	std::shared_ptr<ISound> sound;
 
-	std::shared_ptr<ISound> file = std::shared_ptr<ISound>(new File(filename));
+	std::shared_ptr<ISound> file = std::shared_ptr<ISound>(new File(filename, stream));
 
 	int position = 0;
 
@@ -247,7 +245,7 @@ AUD_API int AUD_readSound(AUD_Sound* sound, float* buffer, int length, int sampl
 
 		buffer[i * 3] = min;
 		buffer[i * 3 + 1] = max;
-		buffer[i * 3 + 2] = sqrt(power / len); // RMS
+		buffer[i * 3 + 2] = std::sqrt(power / len);
 
 		if(overallmax < max)
 			overallmax = max;

extern/audaspace/bindings/C/AUD_Special.h

@@ -37,7 +37,7 @@ extern AUD_API float* AUD_readSoundBuffer(const char* filename, float low, float
 								  float attack, float release, float threshold,
 								  int accumulate, int additive, int square,
 								  float sthreshold, double samplerate,
-								  int* length);
+								  int* length, int stream);
 
 /**
  * Pauses a playing sound after a specific amount of time.

extern/audaspace/bindings/C/AUD_Types.h

@@ -176,5 +176,17 @@ typedef struct
 {
 	AUD_Specs specs;
 	float length;
-	double start_offset;
 } AUD_SoundInfo;
+
+/// Specification of a sound source.
+typedef struct
+{
+	/// Start time in seconds.
+	double start;
+
+	/// Duration in seconds. May be estimated or 0 if unknown.
+	double duration;
+
+	/// Audio data parameters.
+	AUD_DeviceSpecs specs;
+} AUD_StreamInfo;

extern/audaspace/bindings/python/PySound.cpp

@@ -89,10 +89,11 @@ Sound_new(PyTypeObject* type, PyObject* args, PyObject* kwds)
 	self = (Sound*)type->tp_alloc(type, 0);
 	if(self != nullptr)
 	{
-		static const char* kwlist[] = {"filename", nullptr};
+		static const char* kwlist[] = {"filename", "stream", nullptr};
 		const char* filename = nullptr;
+		int stream = 0;
 
-		if(!PyArg_ParseTupleAndKeywords(args, kwds, "s:Sound", const_cast<char**>(kwlist), &filename))
+		if(!PyArg_ParseTupleAndKeywords(args, kwds, "s|i:Sound", const_cast<char**>(kwlist), &filename, &stream))
 		{
 			Py_DECREF(self);
 			return nullptr;
@@ -100,7 +101,7 @@ Sound_new(PyTypeObject* type, PyObject* args, PyObject* kwds)
 
 		try
 		{
-			self->sound = new std::shared_ptr<ISound>(new File(filename));
+			self->sound = new std::shared_ptr<ISound>(new File(filename, stream));
 		}
 		catch(Exception& e)
 		{
@@ -407,8 +408,9 @@ static PyObject *
 Sound_file(PyTypeObject* type, PyObject* args)
 {
 	const char* filename = nullptr;
+	int stream = 0;
 
-	if(!PyArg_ParseTuple(args, "s:file", &filename))
+	if(!PyArg_ParseTuple(args, "s|i:file", &filename, &stream))
 		return nullptr;
 
 	Sound* self;
@@ -418,7 +420,7 @@ Sound_file(PyTypeObject* type, PyObject* args)
 	{
 		try
 		{
-			self->sound = new std::shared_ptr<ISound>(new File(filename));
+			self->sound = new std::shared_ptr<ISound>(new File(filename, stream));
 		}
 		catch(Exception& e)
 		{

extern/audaspace/include/IReader.h

@@ -70,12 +70,6 @@ public:
 	 */
 	virtual int getPosition() const=0;
 
-	/**
-	 * Returns the start offset the sound should have to line up with related sources.
-	 * \return The required start offset in seconds.
-	 */
-	virtual double getStartOffset() const { return 0.0;}
-
 	/**
 	 * Returns the specification of the reader.
 	 * \return The Specs structure.

extern/audaspace/include/file/File.h

@@ -23,9 +23,11 @@
  */
 
 #include "ISound.h"
+#include "FileInfo.h"
 
 #include <string>
 #include <memory>
+#include <vector>
 
 AUD_NAMESPACE_BEGIN
 
@@ -48,6 +50,14 @@ private:
 	 */
 	std::shared_ptr<Buffer> m_buffer;
 
+	/**
+	 * The index of the stream within the file if it contains multiple.
+	 * The first audio stream in the file has index 0 and the index increments by one
+	 * for every other audio stream in the file. Other types of streams in the file
+	 * do not count.
+	 */
+	int m_stream;
+
 	// delete copy constructor and operator=
 	File(const File&) = delete;
 	File& operator=(const File&) = delete;
@@ -57,16 +67,25 @@ public:
 	 * Creates a new sound.
 	 * The file is read from the file system using the given path.
 	 * \param filename The sound file path.
+	 * \param stream The index of the audio stream within the file if it contains multiple audio streams.
 	 */
-	File(std::string filename);
+	File(std::string filename, int stream = 0);
 
 	/**
 	 * Creates a new sound.
 	 * The file is read from memory using the supplied buffer.
 	 * \param buffer The buffer to read from.
 	 * \param size The size of the buffer.
+	 * \param stream The index of the audio stream within the file if it contains multiple audio streams.
 	 */
-	File(const data_t* buffer, int size);
+	File(const data_t* buffer, int size, int stream = 0);
+
+	/**
+	 * Queries the streams of the file.
+	 * \return A vector with as many streams as there are in the file.
+	 * \exception Exception Thrown if the file specified cannot be read.
+	 */
+	std::vector<StreamInfo> queryStreams();
 
 	virtual std::shared_ptr<IReader> createReader();
 };

extern/audaspace/include/file/FileInfo.h

@@ -0,0 +1,42 @@
+/*******************************************************************************
+ * Copyright 2009-2016 Jörg Müller
+ *
+ * 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
+
+/**
+ * @file FileInfo.h
+ * @ingroup file
+ * The FileInfo data structures.
+ */
+
+#include "respec/Specification.h"
+
+AUD_NAMESPACE_BEGIN
+
+/// Specification of a sound source.
+struct StreamInfo
+{
+	/// Start time in seconds.
+	double start;
+
+	/// Duration in seconds. May be estimated or 0 if unknown.
+	double duration;
+
+	/// Audio data parameters.
+	DeviceSpecs specs;
+};
+
+AUD_NAMESPACE_END

extern/audaspace/include/file/FileManager.h

@@ -22,12 +22,14 @@
  * The FileManager class.
  */
 
+#include "FileInfo.h"
 #include "respec/Specification.h"
 #include "IWriter.h"
 
 #include <list>
 #include <memory>
 #include <string>
+#include <vector>
 
 AUD_NAMESPACE_BEGIN
 
@@ -66,18 +68,36 @@ public:
 	/**
 	 * Creates a file reader for the given filename if a registed IFileInput is able to read it.
 	 * @param filename The path to the file.
+	 * @param stream The index of the audio stream within the file if it contains multiple audio streams.
 	 * @return The reader created.
 	 * @exception Exception If no file input can read the file an exception is thrown.
 	 */
-	static std::shared_ptr<IReader> createReader(std::string filename);
+	static std::shared_ptr<IReader> createReader(std::string filename, int stream = 0);
 
 	/**
 	 * Creates a file reader for the given buffer if a registed IFileInput is able to read it.
 	 * @param buffer The buffer to read the file from.
+	 * @param stream The index of the audio stream within the file if it contains multiple audio streams.
 	 * @return The reader created.
 	 * @exception Exception If no file input can read the file an exception is thrown.
 	 */
-	static std::shared_ptr<IReader> createReader(std::shared_ptr<Buffer> buffer);
+	static std::shared_ptr<IReader> createReader(std::shared_ptr<Buffer> buffer, int stream = 0);
+
+	/**
+	 * Queries the streams of a sound file.
+	 * \param filename Path to the file to be read.
+	 * \return A vector with as many streams as there are in the file.
+	 * \exception Exception Thrown if the file specified cannot be read.
+	 */
+	static std::vector<StreamInfo> queryStreams(std::string filename);
+
+	/**
+	 * Queries the streams of a sound file.
+	 * \param buffer The in-memory file buffer.
+	 * \return A vector with as many streams as there are in the file.
+	 * \exception Exception Thrown if the file specified cannot be read.
+	 */
+	static std::vector<StreamInfo> queryStreams(std::shared_ptr<Buffer> buffer);
 
 	/**
 	 * Creates a file writer that writes a sound to the given file path.

extern/audaspace/include/file/IFileInput.h

@@ -23,9 +23,11 @@
  */
 
 #include "Audaspace.h"
+#include "FileInfo.h"
 
 #include <memory>
 #include <string>
+#include <vector>
 
 AUD_NAMESPACE_BEGIN
 
@@ -48,18 +50,36 @@ public:
 	/**
 	 * Creates a reader for a file to be read.
 	 * \param filename Path to the file to be read.
+	 * \param stream The index of the audio stream within the file if it contains multiple audio streams.
 	 * \return The reader that reads the file.
 	 * \exception Exception Thrown if the file specified cannot be read.
 	 */
-	virtual std::shared_ptr<IReader> createReader(std::string filename)=0;
+	virtual std::shared_ptr<IReader> createReader(std::string filename, int stream = 0)=0;
 
 	/**
 	 * Creates a reader for a file to be read from memory.
 	 * \param buffer The in-memory file buffer.
+	 * \param stream The index of the audio stream within the file if it contains multiple audio streams.
 	 * \return The reader that reads the file.
 	 * \exception Exception Thrown if the file specified cannot be read.
 	 */
-	virtual std::shared_ptr<IReader> createReader(std::shared_ptr<Buffer> buffer)=0;
+	virtual std::shared_ptr<IReader> createReader(std::shared_ptr<Buffer> buffer, int stream = 0)=0;
+
+	/**
+	 * Queries the streams of a sound file.
+	 * \param filename Path to the file to be read.
+	 * \return A vector with as many streams as there are in the file.
+	 * \exception Exception Thrown if the file specified cannot be read.
+	 */
+	virtual std::vector<StreamInfo> queryStreams(std::string filename)=0;
+
+	/**
+	 * Queries the streams of a sound file.
+	 * \param buffer The in-memory file buffer.
+	 * \return A vector with as many streams as there are in the file.
+	 * \exception Exception Thrown if the file specified cannot be read.
+	 */
+	virtual std::vector<StreamInfo> queryStreams(std::shared_ptr<Buffer> buffer)=0;
 };
 
 AUD_NAMESPACE_END

extern/audaspace/include/util/RingBuffer.h

@@ -0,0 +1,97 @@
+/*******************************************************************************
+ * Copyright 2009-2021 Jörg Müller
+ *
+ * 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
+
+/**
+ * @file RingBuffer.h
+ * @ingroup util
+ * The RingBuffer class.
+ */
+
+#include "Audaspace.h"
+#include "Buffer.h"
+
+#include <cstddef>
+
+AUD_NAMESPACE_BEGIN
+
+/**
+ * This class is a simple ring buffer in RAM which is 32 Byte aligned and provides
+ * functionality for concurrent reading and writting without locks.
+ */
+class AUD_API RingBuffer
+{
+private:
+	/// The buffer storing the actual data.
+	Buffer m_buffer;
+
+	/// The reading pointer.
+	volatile size_t m_read;
+
+	/// The writing pointer.
+	volatile size_t m_write;
+
+	// delete copy constructor and operator=
+	RingBuffer(const RingBuffer&) = delete;
+	RingBuffer& operator=(const RingBuffer&) = delete;
+
+public:
+	/**
+	 * Creates a new ring buffer.
+	 * \param size The size of the buffer in bytes.
+	 */
+	RingBuffer(int size = 0);
+
+	/**
+	 * Returns the pointer to the ring buffer in memory.
+	 */
+	sample_t* getBuffer() const;
+
+	/**
+	 * Returns the size of the ring buffer in bytes.
+	 */
+	int getSize() const;
+
+	size_t getReadSize() const;
+
+	size_t getWriteSize() const;
+
+	size_t read(data_t* target, size_t size);
+
+	size_t write(data_t* source, size_t size);
+
+	/**
+	 * Resets the ring buffer to a state where nothing has been written or read.
+	 */
+	void reset();
+
+	/**
+	 * Resizes the ring buffer.
+	 * \param size The new size of the ring buffer, measured in bytes.
+	 */
+	void resize(int size);
+
+	/**
+	 * Makes sure the ring buffer has a minimum size.
+	 * If size is >= current size, nothing will happen.
+	 * Otherwise the ring buffer is resized with keep as parameter.
+	 * \param size The new minimum size of the ring buffer, measured in bytes.
+	 */
+	void assureSize(int size);
+};
+
+AUD_NAMESPACE_END

extern/audaspace/plugins/ffmpeg/FFMPEG.cpp

@@ -35,14 +35,24 @@ void FFMPEG::registerPlugin()
 	FileManager::registerOutput(plugin);
 }
 
-std::shared_ptr<IReader> FFMPEG::createReader(std::string filename)
+std::shared_ptr<IReader> FFMPEG::createReader(std::string filename, int stream)
 {
-	return std::shared_ptr<IReader>(new FFMPEGReader(filename));
+	return std::shared_ptr<IReader>(new FFMPEGReader(filename, stream));
 }
 
-std::shared_ptr<IReader> FFMPEG::createReader(std::shared_ptr<Buffer> buffer)
+std::shared_ptr<IReader> FFMPEG::createReader(std::shared_ptr<Buffer> buffer, int stream)
 {
-	return std::shared_ptr<IReader>(new FFMPEGReader(buffer));
+	return std::shared_ptr<IReader>(new FFMPEGReader(buffer, stream));
+}
+
+std::vector<StreamInfo> FFMPEG::queryStreams(std::string filename)
+{
+	return FFMPEGReader(filename).queryStreams();
+}
+
+std::vector<StreamInfo> FFMPEG::queryStreams(std::shared_ptr<Buffer> buffer)
+{
+	return FFMPEGReader(buffer).queryStreams();
 }
 
 std::shared_ptr<IWriter> FFMPEG::createWriter(std::string filename, DeviceSpecs specs, Container format, Codec codec, unsigned int bitrate)

extern/audaspace/plugins/ffmpeg/FFMPEG.h

@@ -52,8 +52,10 @@ public:
 	 */
 	static void registerPlugin();
 
-	virtual std::shared_ptr<IReader> createReader(std::string filename);
-	virtual std::shared_ptr<IReader> createReader(std::shared_ptr<Buffer> buffer);
+	virtual std::shared_ptr<IReader> createReader(std::string filename, int stream = 0);
+	virtual std::shared_ptr<IReader> createReader(std::shared_ptr<Buffer> buffer, int stream = 0);
+	virtual std::vector<StreamInfo> queryStreams(std::string filename);
+	virtual std::vector<StreamInfo> queryStreams(std::shared_ptr<Buffer> buffer);
 	virtual std::shared_ptr<IWriter> createWriter(std::string filename, DeviceSpecs specs, Container format, Codec codec, unsigned int bitrate);
 };
 

extern/audaspace/plugins/ffmpeg/FFMPEGReader.cpp

@@ -31,6 +31,25 @@ AUD_NAMESPACE_BEGIN
 #define FFMPEG_OLD_CODE
 #endif
 
+SampleFormat FFMPEGReader::convertSampleFormat(AVSampleFormat format)
+{
+	switch(av_get_packed_sample_fmt(format))
+	{
+	case AV_SAMPLE_FMT_U8:
+		return FORMAT_U8;
+	case AV_SAMPLE_FMT_S16:
+		return FORMAT_S16;
+	case AV_SAMPLE_FMT_S32:
+		return FORMAT_S32;
+	case AV_SAMPLE_FMT_FLT:
+		return FORMAT_FLOAT32;
+	case AV_SAMPLE_FMT_DBL:
+		return FORMAT_FLOAT64;
+	default:
+		AUD_THROW(FileException, "FFMPEG sample format unknown.");
+	}
+}
+
 int FFMPEGReader::decode(AVPacket& packet, Buffer& buffer)
 {
 	int buf_size = buffer.getSize();
@@ -68,7 +87,7 @@ int FFMPEGReader::decode(AVPacket& packet, Buffer& buffer)
 					for(int i = 0; i < m_frame->nb_samples; i++)
 					{
 						std::memcpy(((data_t*)buffer.getBuffer()) + buf_pos + ((m_codecCtx->channels * i) + channel) * single_size,
-								 m_frame->data[channel] + i * single_size, single_size);
+							   m_frame->data[channel] + i * single_size, single_size);
 					}
 				}
 			}
@@ -109,7 +128,7 @@ int FFMPEGReader::decode(AVPacket& packet, Buffer& buffer)
 				for(int i = 0; i < m_frame->nb_samples; i++)
 				{
 					std::memcpy(((data_t*)buffer.getBuffer()) + buf_pos + ((m_codecCtx->channels * i) + channel) * single_size,
-							 m_frame->data[channel] + i * single_size, single_size);
+						   m_frame->data[channel] + i * single_size, single_size);
 				}
 			}
 		}
@@ -123,13 +142,10 @@ int FFMPEGReader::decode(AVPacket& packet, Buffer& buffer)
 	return buf_pos;
 }
 
-void FFMPEGReader::init()
+void FFMPEGReader::init(int stream)
 {
 	m_position = 0;
-	m_start_offset = 0.0f;
 	m_pkgbuf_left = 0;
-	m_st_time = 0;
-	m_duration = 0;
 
 	if(avformat_find_stream_info(m_formatCtx, nullptr) < 0)
 		AUD_THROW(FileException, "File couldn't be read, ffmpeg couldn't find the stream info.");
@@ -137,43 +153,22 @@ void FFMPEGReader::init()
 	// find audio stream and codec
 	m_stream = -1;
 
-	double dur_sec = 0;
-
 	for(unsigned int i = 0; i < m_formatCtx->nb_streams; i++)
 	{
 #ifdef FFMPEG_OLD_CODE
-		if(m_formatCtx->streams[i]->codec->codec_type == AVMEDIA_TYPE_AUDIO)
+		if((m_formatCtx->streams[i]->codec->codec_type == AVMEDIA_TYPE_AUDIO)
 #else
-		if(m_formatCtx->streams[i]->codecpar->codec_type == AVMEDIA_TYPE_AUDIO)
+		if((m_formatCtx->streams[i]->codecpar->codec_type == AVMEDIA_TYPE_AUDIO)
 #endif
+			&& (m_stream < 0))
 		{
-			AVStream *audio_stream = m_formatCtx->streams[i];
-			double audio_timebase = av_q2d(audio_stream->time_base);
-
-			if (audio_stream->start_time != AV_NOPTS_VALUE)
+			if(stream == 0)
 			{
-				m_st_time = audio_stream->start_time;
-			}
-
-			int64_t ctx_start_time = 0;
-			if (m_formatCtx->start_time != AV_NOPTS_VALUE) {
-				ctx_start_time = m_formatCtx->start_time;
-			}
-
-			m_start_offset = m_st_time * audio_timebase - (double)ctx_start_time / AV_TIME_BASE;
-
-			if(audio_stream->duration != AV_NOPTS_VALUE)
-			{
-				dur_sec = audio_stream->duration * audio_timebase;
+				m_stream=i;
+				break;
 			}
 			else
-			{
-				/* If the audio starts after the stream start time, subract this from the total duration. */
-				dur_sec = (double)m_formatCtx->duration / AV_TIME_BASE - m_start_offset;
-			}
-
-			m_stream=i;
-			break;
+				stream--;
 		}
 	}
 
@@ -242,10 +237,9 @@ void FFMPEGReader::init()
 	}
 
 	m_specs.rate = (SampleRate) m_codecCtx->sample_rate;
-	m_duration = lround(dur_sec * m_codecCtx->sample_rate);
 }
 
-FFMPEGReader::FFMPEGReader(std::string filename) :
+FFMPEGReader::FFMPEGReader(std::string filename, int stream) :
 	m_pkgbuf(),
 	m_formatCtx(nullptr),
 	m_codecCtx(nullptr),
@@ -259,7 +253,7 @@ FFMPEGReader::FFMPEGReader(std::string filename) :
 
 	try
 	{
-		init();
+		init(stream);
 	}
 	catch(Exception&)
 	{
@@ -268,7 +262,7 @@ FFMPEGReader::FFMPEGReader(std::string filename) :
 	}
 }
 
-FFMPEGReader::FFMPEGReader(std::shared_ptr<Buffer> buffer) :
+FFMPEGReader::FFMPEGReader(std::shared_ptr<Buffer> buffer, int stream) :
 		m_pkgbuf(),
 		m_codecCtx(nullptr),
 		m_frame(nullptr),
@@ -295,7 +289,7 @@ FFMPEGReader::FFMPEGReader(std::shared_ptr<Buffer> buffer) :
 
 	try
 	{
-		init();
+		init(stream);
 	}
 	catch(Exception&)
 	{
@@ -318,6 +312,51 @@ FFMPEGReader::~FFMPEGReader()
 	avformat_close_input(&m_formatCtx);
 }
 
+std::vector<StreamInfo> FFMPEGReader::queryStreams()
+{
+	std::vector<StreamInfo> result;
+
+	for(unsigned int i = 0; i < m_formatCtx->nb_streams; i++)
+	{
+#ifdef FFMPEG_OLD_CODE
+		if(m_formatCtx->streams[i]->codec->codec_type == AVMEDIA_TYPE_AUDIO)
+#else
+		if(m_formatCtx->streams[i]->codecpar->codec_type == AVMEDIA_TYPE_AUDIO)
+#endif
+		{
+			StreamInfo info;
+
+			double time_base = av_q2d(m_formatCtx->streams[i]->time_base);
+
+			if(m_formatCtx->streams[i]->start_time != AV_NOPTS_VALUE)
+				info.start = m_formatCtx->streams[i]->start_time * time_base;
+			else
+				info.start = 0;
+
+			if(m_formatCtx->streams[i]->duration != AV_NOPTS_VALUE)
+				info.duration = m_formatCtx->streams[i]->duration * time_base;
+			else if(m_formatCtx->duration != AV_NOPTS_VALUE)
+				info.duration = double(m_formatCtx->duration) / AV_TIME_BASE - info.start;
+			else
+				info.duration = 0;
+
+#ifdef FFMPEG_OLD_CODE
+			info.specs.channels = Channels(m_formatCtx->streams[i]->codec->channels);
+			info.specs.rate = m_formatCtx->streams[i]->codec->sample_rate;
+			info.specs.format = convertSampleFormat(m_formatCtx->streams[i]->codec->sample_fmt);
+#else
+			info.specs.channels = Channels(m_formatCtx->streams[i]->codecpar->channels);
+			info.specs.rate = m_formatCtx->streams[i]->codecpar->sample_rate;
+			info.specs.format = convertSampleFormat(AVSampleFormat(m_formatCtx->streams[i]->codecpar->format));
+#endif
+
+			result.emplace_back(info);
+		}
+	}
+
+	return result;
+}
+
 int FFMPEGReader::read_packet(void* opaque, uint8_t* buf, int buf_size)
 {
 	FFMPEGReader* reader = reinterpret_cast<FFMPEGReader*>(opaque);
@@ -368,18 +407,16 @@ void FFMPEGReader::seek(int position)
 {
 	if(position >= 0)
 	{
-		double pts_time_base =
-			av_q2d(m_formatCtx->streams[m_stream]->time_base);
+		double pts_time_base = av_q2d(m_formatCtx->streams[m_stream]->time_base);
 
-		uint64_t seek_pts = (((uint64_t)position) / ((uint64_t)m_specs.rate)) / pts_time_base;
+		uint64_t st_time = m_formatCtx->streams[m_stream]->start_time;
+		uint64_t seek_pos = (uint64_t)(position / (pts_time_base * m_specs.rate));
 
-		if(m_st_time != AV_NOPTS_VALUE) {
-			seek_pts += m_st_time;
-		}
+		if(st_time != AV_NOPTS_VALUE)
+			seek_pos += st_time;
 
 		// a value < 0 tells us that seeking failed
-		if(av_seek_frame(m_formatCtx, m_stream, seek_pts,
-				 AVSEEK_FLAG_BACKWARD | AVSEEK_FLAG_ANY) >= 0)
+		if(av_seek_frame(m_formatCtx, m_stream, seek_pos, AVSEEK_FLAG_BACKWARD | AVSEEK_FLAG_ANY) >= 0)
 		{
 			avcodec_flush_buffers(m_codecCtx);
 			m_position = position;
@@ -400,7 +437,7 @@ void FFMPEGReader::seek(int position)
 					if(packet.pts != AV_NOPTS_VALUE)
 					{
 						// calculate real position, and read to frame!
-						m_position = (packet.pts - m_st_time) * pts_time_base * m_specs.rate;
+						m_position = (packet.pts - (st_time != AV_NOPTS_VALUE ? st_time : 0)) * pts_time_base * m_specs.rate;
 
 						if(m_position < position)
 						{
@@ -430,8 +467,25 @@ void FFMPEGReader::seek(int position)
 
 int FFMPEGReader::getLength() const
 {
+	auto stream = m_formatCtx->streams[m_stream];
+
+	double time_base = av_q2d(stream->time_base);
+	double duration;
+
+	if(stream->duration != AV_NOPTS_VALUE)
+		duration = stream->duration * time_base;
+	else if(m_formatCtx->duration != AV_NOPTS_VALUE)
+	{
+		duration = float(m_formatCtx->duration) / AV_TIME_BASE;
+
+		if(stream->start_time != AV_NOPTS_VALUE)
+			duration -= stream->start_time * time_base;
+	}
+	else
+		duration = -1;
+
 	// return approximated remaning size
-	return m_duration - m_position;
+	return (int)(duration * m_codecCtx->sample_rate) - m_position;
 }
 
 int FFMPEGReader::getPosition() const
@@ -439,11 +493,6 @@ int FFMPEGReader::getPosition() const
 	return m_position;
 }
 
-double FFMPEGReader::getStartOffset() const
-{
-	return m_start_offset;
-}
-
 Specs FFMPEGReader::getSpecs() const
 {
 	return m_specs.specs;
@@ -480,13 +529,11 @@ void FFMPEGReader::read(int& length, bool& eos, sample_t* buffer)
 			// decode the package
 			pkgbuf_pos = decode(packet, m_pkgbuf);
 
-			if (packet.pts >= m_st_time) {
-				// copy to output buffer
-				data_size = std::min(pkgbuf_pos, left * sample_size);
-				m_convert((data_t*) buf, (data_t*) m_pkgbuf.getBuffer(), data_size / AUD_FORMAT_SIZE(m_specs.format));
-				buf += data_size / AUD_FORMAT_SIZE(m_specs.format);
-				left -= data_size / sample_size;
-			}
+			// copy to output buffer
+			data_size = std::min(pkgbuf_pos, left * sample_size);
+			m_convert((data_t*) buf, (data_t*) m_pkgbuf.getBuffer(), data_size / AUD_FORMAT_SIZE(m_specs.format));
+			buf += data_size / AUD_FORMAT_SIZE(m_specs.format);
+			left -= data_size / sample_size;
 		}
 		av_packet_unref(&packet);
 	}

extern/audaspace/plugins/ffmpeg/FFMPEGReader.h

@@ -29,9 +29,11 @@
 #include "respec/ConverterFunctions.h"
 #include "IReader.h"
 #include "util/Buffer.h"
+#include "file/FileInfo.h"
 
 #include <string>
 #include <memory>
+#include <vector>
 
 struct AVCodecContext;
 extern "C" {
@@ -54,22 +56,6 @@ private:
 	 */
 	int m_position;
 
-	/**
-	 * The start offset in seconds relative to the media container start time.
-	 * IE how much the sound should be delayed to be kept in sync with the rest of the containter streams.
-	 */
-	double m_start_offset;
-
-	/**
-	 * The start time pts of the stream. All packets before this timestamp shouldn't be played back (only decoded).
-	 */
-	int64_t m_st_time;
-
-	/**
-	 * The duration of the audio stream in samples.
-	 */
-	int64_t m_duration;
-
 	/**
 	 * The specification of the audio data.
 	 */
@@ -135,6 +121,13 @@ private:
 	 */
 	bool m_tointerleave;
 
+	/**
+	 * Converts an ffmpeg sample format to an audaspace one.
+	 * \param format The AVSampleFormat sample format.
+	 * \return The sample format as SampleFormat.
+	 */
+	AUD_LOCAL static SampleFormat convertSampleFormat(AVSampleFormat format);
+
 	/**
 	 * Decodes a packet into the given buffer.
 	 * \param packet The AVPacket to decode.
@@ -145,8 +138,9 @@ private:
 
 	/**
 	 * Initializes the object.
+	 * \param stream The index of the audio stream within the file if it contains multiple audio streams.
 	 */
-	AUD_LOCAL void init();
+	AUD_LOCAL void init(int stream);
 
 	// delete copy constructor and operator=
 	FFMPEGReader(const FFMPEGReader&) = delete;
@@ -156,24 +150,33 @@ public:
 	/**
 	 * Creates a new reader.
 	 * \param filename The path to the file to be read.
+	 * \param stream The index of the audio stream within the file if it contains multiple audio streams.
 	 * \exception Exception Thrown if the file specified does not exist or
 	 *            cannot be read with ffmpeg.
 	 */
-	FFMPEGReader(std::string filename);
+	FFMPEGReader(std::string filename, int stream = 0);
 
 	/**
 	 * Creates a new reader.
 	 * \param buffer The buffer to read from.
+	 * \param stream The index of the audio stream within the file if it contains multiple audio streams.
 	 * \exception Exception Thrown if the buffer specified cannot be read
 	 *                          with ffmpeg.
 	 */
-	FFMPEGReader(std::shared_ptr<Buffer> buffer);
+	FFMPEGReader(std::shared_ptr<Buffer> buffer, int stream = 0);
 
 	/**
 	 * Destroys the reader and closes the file.
 	 */
 	virtual ~FFMPEGReader();
 
+	/**
+	 * Queries the streams of a sound file.
+	 * \return A vector with as many streams as there are in the file.
+	 * \exception Exception Thrown if the file specified cannot be read.
+	 */
+	virtual std::vector<StreamInfo> queryStreams();
+
 	/**
 	 * Reads data to a memory buffer.
 	 * This function is used for avio only.
@@ -198,7 +201,6 @@ public:
 	virtual void seek(int position);
 	virtual int getLength() const;
 	virtual int getPosition() const;
-	virtual double getStartOffset() const;
 	virtual Specs getSpecs() const;
 	virtual void read(int& length, bool& eos, sample_t* buffer);
 };

extern/audaspace/plugins/libsndfile/SndFile.cpp

@@ -32,16 +32,26 @@ void SndFile::registerPlugin()
 	FileManager::registerOutput(plugin);
 }
 
-std::shared_ptr<IReader> SndFile::createReader(std::string filename)
+std::shared_ptr<IReader> SndFile::createReader(std::string filename, int stream)
 {
 	return std::shared_ptr<IReader>(new SndFileReader(filename));
 }
 
-std::shared_ptr<IReader> SndFile::createReader(std::shared_ptr<Buffer> buffer)
+std::shared_ptr<IReader> SndFile::createReader(std::shared_ptr<Buffer> buffer, int stream)
 {
 	return std::shared_ptr<IReader>(new SndFileReader(buffer));
 }
 
+std::vector<StreamInfo> SndFile::queryStreams(std::string filename)
+{
+	return SndFileReader(filename).queryStreams();
+}
+
+std::vector<StreamInfo> SndFile::queryStreams(std::shared_ptr<Buffer> buffer)
+{
+	return SndFileReader(buffer).queryStreams();
+}
+
 std::shared_ptr<IWriter> SndFile::createWriter(std::string filename, DeviceSpecs specs, Container format, Codec codec, unsigned int bitrate)
 {
 	return std::shared_ptr<IWriter>(new SndFileWriter(filename, specs, format, codec, bitrate));

extern/audaspace/plugins/libsndfile/SndFile.h

@@ -52,8 +52,10 @@ public:
 	 */
 	static void registerPlugin();
 
-	virtual std::shared_ptr<IReader> createReader(std::string filename);
-	virtual std::shared_ptr<IReader> createReader(std::shared_ptr<Buffer> buffer);
+	virtual std::shared_ptr<IReader> createReader(std::string filename, int stream = 0);
+	virtual std::shared_ptr<IReader> createReader(std::shared_ptr<Buffer> buffer, int stream = 0);
+	virtual std::vector<StreamInfo> queryStreams(std::string filename);
+	virtual std::vector<StreamInfo> queryStreams(std::shared_ptr<Buffer> buffer);
 	virtual std::shared_ptr<IWriter> createWriter(std::string filename, DeviceSpecs specs, Container format, Codec codec, unsigned int bitrate);
 };
 

extern/audaspace/plugins/libsndfile/SndFileReader.cpp

@@ -118,6 +118,21 @@ SndFileReader::~SndFileReader()
 	sf_close(m_sndfile);
 }
 
+std::vector<StreamInfo> SndFileReader::queryStreams()
+{
+	std::vector<StreamInfo> result;
+
+	StreamInfo info;
+	info.start = 0;
+	info.duration = double(getLength()) / m_specs.rate;
+	info.specs.specs = m_specs;
+	info.specs.format = FORMAT_FLOAT32;
+
+	result.emplace_back(info);
+
+	return result;
+}
+
 bool SndFileReader::isSeekable() const
 {
 	return m_seekable;

extern/audaspace/plugins/libsndfile/SndFileReader.h

@@ -28,9 +28,12 @@
  * The SndFileReader class.
  */
 
+#include "file/FileInfo.h"
+
 #include <string>
 #include <sndfile.h>
 #include <memory>
+#include <vector>
 
 AUD_NAMESPACE_BEGIN
 
@@ -96,6 +99,7 @@ public:
 	/**
 	 * Creates a new reader.
 	 * \param filename The path to the file to be read.
+	 * \param stream The index of the audio stream within the file if it contains multiple audio streams.
 	 * \exception Exception Thrown if the file specified does not exist or
 	 *            cannot be read with libsndfile.
 	 */
@@ -104,6 +108,7 @@ public:
 	/**
 	 * Creates a new reader.
 	 * \param buffer The buffer to read from.
+	 * \param stream The index of the audio stream within the file if it contains multiple audio streams.
 	 * \exception Exception Thrown if the buffer specified cannot be read
 	 *                          with libsndfile.
 	 */
@@ -114,6 +119,13 @@ public:
 	 */
 	virtual ~SndFileReader();
 
+	/**
+	 * Queries the streams of a sound file.
+	 * \return A vector with as many streams as there are in the file.
+	 * \exception Exception Thrown if the file specified cannot be read.
+	 */
+	virtual std::vector<StreamInfo> queryStreams();
+
 	virtual bool isSeekable() const;
 	virtual void seek(int position);
 	virtual int getLength() const;

extern/audaspace/plugins/pulseaudio/PulseAudioDevice.cpp

@@ -23,95 +23,121 @@
 
 AUD_NAMESPACE_BEGIN
 
+PulseAudioDevice::PulseAudioSynchronizer::PulseAudioSynchronizer(PulseAudioDevice *device) :
+	m_device(device)
+{
+}
+
+double PulseAudioDevice::PulseAudioSynchronizer::getPosition(std::shared_ptr<IHandle> handle)
+{
+	pa_usec_t latency;
+	int negative;
+	AUD_pa_stream_get_latency(m_device->m_stream, &latency, &negative);
+
+	double delay = m_device->m_ring_buffer.getReadSize() / (AUD_SAMPLE_SIZE(m_device->m_specs) * m_device->m_specs.rate) + latency * 1.0e-6;
+
+	return handle->getPosition() - delay;
+}
+
+void PulseAudioDevice::updateRingBuffer()
+{
+	unsigned int samplesize = AUD_SAMPLE_SIZE(m_specs);
+
+	std::unique_lock<std::mutex> lock(m_mixingLock);
+
+	Buffer buffer;
+
+	while(m_valid)
+	{
+		size_t size = m_ring_buffer.getWriteSize();
+
+		size_t sample_count = size / samplesize;
+
+		if(sample_count > 0)
+		{
+			size = sample_count * samplesize;
+
+			buffer.assureSize(size);
+
+			mix(reinterpret_cast<data_t*>(buffer.getBuffer()), sample_count);
+
+			m_ring_buffer.write(reinterpret_cast<data_t*>(buffer.getBuffer()), size);
+		}
+
+		m_mixingCondition.wait(lock);
+	}
+}
+
 void PulseAudioDevice::PulseAudio_state_callback(pa_context *context, void *data)
 {
 	PulseAudioDevice* device = (PulseAudioDevice*)data;
 
-	std::lock_guard<ILockable> lock(*device);
-
 	device->m_state = AUD_pa_context_get_state(context);
+
+	AUD_pa_threaded_mainloop_signal(device->m_mainloop, 0);
 }
 
 void PulseAudioDevice::PulseAudio_request(pa_stream *stream, size_t total_bytes, void *data)
 {
 	PulseAudioDevice* device = (PulseAudioDevice*)data;
 
-	void* buffer;
+	data_t* buffer;
+
+	size_t sample_size = AUD_DEVICE_SAMPLE_SIZE(device->m_specs);
 
 	while(total_bytes > 0)
 	{
 		size_t num_bytes = total_bytes;
 
-		AUD_pa_stream_begin_write(stream, &buffer, &num_bytes);
+		AUD_pa_stream_begin_write(stream, reinterpret_cast<void**>(&buffer), &num_bytes);
 
-		device->mix((data_t*)buffer, num_bytes / AUD_DEVICE_SAMPLE_SIZE(device->m_specs));
+		size_t readsamples = device->m_ring_buffer.getReadSize();
 
-		AUD_pa_stream_write(stream, buffer, num_bytes, nullptr, 0, PA_SEEK_RELATIVE);
+		readsamples = std::min(readsamples, size_t(num_bytes)) / sample_size;
+
+		device->m_ring_buffer.read(buffer, readsamples * sample_size);
+
+		if(readsamples * sample_size < num_bytes)
+			std::memset(buffer + readsamples * sample_size, 0, num_bytes - readsamples * sample_size);
+
+		if(device->m_mixingLock.try_lock())
+		{
+			device->m_mixingCondition.notify_all();
+			device->m_mixingLock.unlock();
+		}
+
+		AUD_pa_stream_write(stream, reinterpret_cast<void*>(buffer), num_bytes, nullptr, 0, PA_SEEK_RELATIVE);
 
 		total_bytes -= num_bytes;
 	}
 }
 
-void PulseAudioDevice::PulseAudio_underflow(pa_stream *stream, void *data)
+void PulseAudioDevice::playing(bool playing)
 {
-	PulseAudioDevice* device = (PulseAudioDevice*)data;
+	m_playback = playing;
 
-	DeviceSpecs specs = device->getSpecs();
-
-	if(++device->m_underflows > 4 && device->m_buffersize < AUD_DEVICE_SAMPLE_SIZE(specs) * specs.rate * 2)
-	{
-		device->m_buffersize <<= 1;
-		device->m_underflows = 0;
-
-		pa_buffer_attr buffer_attr;
-
-		buffer_attr.fragsize = -1U;
-		buffer_attr.maxlength = -1U;
-		buffer_attr.minreq = -1U;
-		buffer_attr.prebuf = -1U;
-		buffer_attr.tlength = device->m_buffersize;
-
-		AUD_pa_stream_set_buffer_attr(stream, &buffer_attr, nullptr, nullptr);
-	}
-}
-
-void PulseAudioDevice::runMixingThread()
-{
-	for(;;)
-	{
-		{
-			std::lock_guard<ILockable> lock(*this);
-
-			if(shouldStop())
-			{
-				AUD_pa_stream_cork(m_stream, 1, nullptr, nullptr);
-				AUD_pa_stream_flush(m_stream, nullptr, nullptr);
-				doStop();
-				return;
-			}
-		}
-
-		if(AUD_pa_stream_is_corked(m_stream))
-			AUD_pa_stream_cork(m_stream, 0, nullptr, nullptr);
-
-		// similar to AUD_pa_mainloop_iterate(m_mainloop, false, nullptr); except with a longer timeout
-		AUD_pa_mainloop_prepare(m_mainloop, 1 << 14);
-		AUD_pa_mainloop_poll(m_mainloop);
-		AUD_pa_mainloop_dispatch(m_mainloop);
-	}
+	AUD_pa_threaded_mainloop_lock(m_mainloop);
+	AUD_pa_stream_cork(m_stream, playing ? 0 : 1, nullptr, nullptr);
+	AUD_pa_threaded_mainloop_unlock(m_mainloop);
 }
 
 PulseAudioDevice::PulseAudioDevice(std::string name, DeviceSpecs specs, int buffersize) :
+	m_synchronizer(this),
+	m_playback(false),
 	m_state(PA_CONTEXT_UNCONNECTED),
+	m_valid(true),
 	m_underflows(0)
 {
-	m_mainloop = AUD_pa_mainloop_new();
+	m_mainloop = AUD_pa_threaded_mainloop_new();
 
-	m_context = AUD_pa_context_new(AUD_pa_mainloop_get_api(m_mainloop), name.c_str());
+	AUD_pa_threaded_mainloop_lock(m_mainloop);
+
+	m_context = AUD_pa_context_new(AUD_pa_threaded_mainloop_get_api(m_mainloop), name.c_str());
 
 	if(!m_context)
 	{
-		AUD_pa_mainloop_free(m_mainloop);
+		AUD_pa_threaded_mainloop_unlock(m_mainloop);
+		AUD_pa_threaded_mainloop_free(m_mainloop);
 
 		AUD_THROW(DeviceException, "Could not connect to PulseAudio.");
 	}
@@ -120,21 +146,26 @@ PulseAudioDevice::PulseAudioDevice(std::string name, DeviceSpecs specs, int buff
 
 	AUD_pa_context_connect(m_context, nullptr, PA_CONTEXT_NOFLAGS, nullptr);
 
+	AUD_pa_threaded_mainloop_start(m_mainloop);
+
 	while(m_state != PA_CONTEXT_READY)
 	{
 		switch(m_state)
 		{
 		case PA_CONTEXT_FAILED:
 		case PA_CONTEXT_TERMINATED:
+			AUD_pa_threaded_mainloop_unlock(m_mainloop);
+			AUD_pa_threaded_mainloop_stop(m_mainloop);
+
 			AUD_pa_context_disconnect(m_context);
 			AUD_pa_context_unref(m_context);
 
-			AUD_pa_mainloop_free(m_mainloop);
+			AUD_pa_threaded_mainloop_free(m_mainloop);
 
 			AUD_THROW(DeviceException, "Could not connect to PulseAudio.");
 			break;
 		default:
-			AUD_pa_mainloop_iterate(m_mainloop, true, nullptr);
+			AUD_pa_threaded_mainloop_wait(m_mainloop);
 			break;
 		}
 	}
@@ -182,16 +213,18 @@ PulseAudioDevice::PulseAudioDevice(std::string name, DeviceSpecs specs, int buff
 
 	if(!m_stream)
 	{
+		AUD_pa_threaded_mainloop_unlock(m_mainloop);
+		AUD_pa_threaded_mainloop_stop(m_mainloop);
+
 		AUD_pa_context_disconnect(m_context);
 		AUD_pa_context_unref(m_context);
 
-		AUD_pa_mainloop_free(m_mainloop);
+		AUD_pa_threaded_mainloop_free(m_mainloop);
 
 		AUD_THROW(DeviceException, "Could not create PulseAudio stream.");
 	}
 
 	AUD_pa_stream_set_write_callback(m_stream, PulseAudio_request, this);
-	AUD_pa_stream_set_underflow_callback(m_stream, PulseAudio_underflow, this);
 
 	buffersize *= AUD_DEVICE_SAMPLE_SIZE(m_specs);
 	m_buffersize = buffersize;
@@ -204,31 +237,53 @@ PulseAudioDevice::PulseAudioDevice(std::string name, DeviceSpecs specs, int buff
 	buffer_attr.prebuf = -1U;
 	buffer_attr.tlength = buffersize;
 
-	if(AUD_pa_stream_connect_playback(m_stream, nullptr, &buffer_attr, static_cast<pa_stream_flags_t>(PA_STREAM_START_CORKED | PA_STREAM_INTERPOLATE_TIMING | PA_STREAM_ADJUST_LATENCY | PA_STREAM_AUTO_TIMING_UPDATE), nullptr, nullptr) < 0)
+	m_ring_buffer.resize(buffersize);
+
+	if(AUD_pa_stream_connect_playback(m_stream, nullptr, &buffer_attr, static_cast<pa_stream_flags_t>(PA_STREAM_INTERPOLATE_TIMING | PA_STREAM_ADJUST_LATENCY | PA_STREAM_AUTO_TIMING_UPDATE), nullptr, nullptr) < 0)
 	{
+		AUD_pa_threaded_mainloop_unlock(m_mainloop);
+		AUD_pa_threaded_mainloop_stop(m_mainloop);
+
 		AUD_pa_context_disconnect(m_context);
 		AUD_pa_context_unref(m_context);
 
-		AUD_pa_mainloop_free(m_mainloop);
+		AUD_pa_threaded_mainloop_free(m_mainloop);
 
 		AUD_THROW(DeviceException, "Could not connect PulseAudio stream.");
 	}
 
+	AUD_pa_threaded_mainloop_unlock(m_mainloop);
+
 	create();
+
+	m_mixingThread = std::thread(&PulseAudioDevice::updateRingBuffer, this);
 }
 
 PulseAudioDevice::~PulseAudioDevice()
 {
-	stopMixingThread();
+	m_valid = false;
+
+	m_mixingLock.lock();
+	m_mixingCondition.notify_all();
+	m_mixingLock.unlock();
+
+	m_mixingThread.join();
+
+	AUD_pa_threaded_mainloop_stop(m_mainloop);
 
 	AUD_pa_context_disconnect(m_context);
 	AUD_pa_context_unref(m_context);
 
-	AUD_pa_mainloop_free(m_mainloop);
+	AUD_pa_threaded_mainloop_free(m_mainloop);
 
 	destroy();
 }
 
+ISynchronizer *PulseAudioDevice::getSynchronizer()
+{
+	return &m_synchronizer;
+}
+
 class PulseAudioDeviceFactory : public IDeviceFactory
 {
 private:

extern/audaspace/plugins/pulseaudio/PulseAudioDevice.h

@@ -26,7 +26,11 @@
  * The PulseAudioDevice class.
  */
 
-#include "devices/ThreadedDevice.h"
+#include "devices/SoftwareDevice.h"
+#include "util/RingBuffer.h"
+
+#include <condition_variable>
+#include <thread>
 
 #include <pulse/pulseaudio.h>
 
@@ -35,17 +39,65 @@ AUD_NAMESPACE_BEGIN
 /**
  * This device plays back through PulseAudio, the simple direct media layer.
  */
-class AUD_PLUGIN_API PulseAudioDevice : public ThreadedDevice
+class AUD_PLUGIN_API PulseAudioDevice : public SoftwareDevice
 {
 private:
-	pa_mainloop* m_mainloop;
+	class PulseAudioSynchronizer : public DefaultSynchronizer
+	{
+		PulseAudioDevice* m_device;
+
+	public:
+		PulseAudioSynchronizer(PulseAudioDevice* device);
+
+		virtual double getPosition(std::shared_ptr<IHandle> handle);
+	};
+
+	/// Synchronizer.
+	PulseAudioSynchronizer m_synchronizer;
+
+	/**
+	 * Whether there is currently playback.
+	 */
+	volatile bool m_playback;
+
+	pa_threaded_mainloop* m_mainloop;
 	pa_context* m_context;
 	pa_stream* m_stream;
 	pa_context_state_t m_state;
 
+	/**
+	 * The mixing ring buffer.
+	 */
+	RingBuffer m_ring_buffer;
+
+	/**
+	 * Whether the device is valid.
+	 */
+	bool m_valid;
+
 	int m_buffersize;
 	uint32_t m_underflows;
 
+	/**
+	 * The mixing thread.
+	 */
+	std::thread m_mixingThread;
+
+	/**
+	 * Mutex for mixing.
+	 */
+	std::mutex m_mixingLock;
+
+	/**
+	 * Condition for mixing.
+	 */
+	std::condition_variable m_mixingCondition;
+
+	/**
+	 * Updates the ring buffer.
+	 */
+	AUD_LOCAL void updateRingBuffer();
+
 	/**
 	 * Reports the state of the PulseAudio server connection.
 	 * \param context The PulseAudio context.
@@ -61,23 +113,13 @@ private:
 	 */
 	AUD_LOCAL static void PulseAudio_request(pa_stream* stream, size_t total_bytes, void* data);
 
-	/**
-	 * Reports an underflow from the PulseAudio server.
-	 * Automatically adjusts the latency if this happens too often.
-	 * @param stream The PulseAudio stream.
-	 * \param data The PulseAudio device.
-	 */
-	AUD_LOCAL static void PulseAudio_underflow(pa_stream* stream, void* data);
-
-	/**
-	 * Streaming thread main function.
-	 */
-	AUD_LOCAL void runMixingThread();
-
 	// delete copy constructor and operator=
 	PulseAudioDevice(const PulseAudioDevice&) = delete;
 	PulseAudioDevice& operator=(const PulseAudioDevice&) = delete;
 
+protected:
+	virtual void playing(bool playing);
+
 public:
 	/**
 	 * Opens the PulseAudio audio device for playback.
@@ -93,6 +135,8 @@ public:
 	 */
 	virtual ~PulseAudioDevice();
 
+	virtual ISynchronizer* getSynchronizer();
+
 	/**
 	 * Registers this plugin.
 	 */

extern/audaspace/plugins/pulseaudio/PulseAudioSymbols.h

@@ -25,6 +25,7 @@ PULSEAUDIO_SYMBOL(pa_stream_begin_write);
 PULSEAUDIO_SYMBOL(pa_stream_connect_playback);
 PULSEAUDIO_SYMBOL(pa_stream_cork);
 PULSEAUDIO_SYMBOL(pa_stream_flush);
+PULSEAUDIO_SYMBOL(pa_stream_get_latency);
 PULSEAUDIO_SYMBOL(pa_stream_is_corked);
 PULSEAUDIO_SYMBOL(pa_stream_new);
 PULSEAUDIO_SYMBOL(pa_stream_set_buffer_attr);
@@ -39,3 +40,13 @@ PULSEAUDIO_SYMBOL(pa_mainloop_iterate);
 PULSEAUDIO_SYMBOL(pa_mainloop_prepare);
 PULSEAUDIO_SYMBOL(pa_mainloop_poll);
 PULSEAUDIO_SYMBOL(pa_mainloop_dispatch);
+
+PULSEAUDIO_SYMBOL(pa_threaded_mainloop_free);
+PULSEAUDIO_SYMBOL(pa_threaded_mainloop_get_api);
+PULSEAUDIO_SYMBOL(pa_threaded_mainloop_lock);
+PULSEAUDIO_SYMBOL(pa_threaded_mainloop_new);
+PULSEAUDIO_SYMBOL(pa_threaded_mainloop_signal);
+PULSEAUDIO_SYMBOL(pa_threaded_mainloop_start);
+PULSEAUDIO_SYMBOL(pa_threaded_mainloop_stop);
+PULSEAUDIO_SYMBOL(pa_threaded_mainloop_unlock);
+PULSEAUDIO_SYMBOL(pa_threaded_mainloop_wait);

extern/audaspace/src/file/File.cpp

@@ -23,23 +23,31 @@
 
 AUD_NAMESPACE_BEGIN
 
-File::File(std::string filename) :
-	m_filename(filename)
+File::File(std::string filename, int stream) :
+	m_filename(filename), m_stream(stream)
 {
 }
 
-File::File(const data_t* buffer, int size) :
-	m_buffer(new Buffer(size))
+File::File(const data_t* buffer, int size, int stream) :
+	m_buffer(new Buffer(size)), m_stream(stream)
 {
 	std::memcpy(m_buffer->getBuffer(), buffer, size);
 }
 
+std::vector<StreamInfo> File::queryStreams()
+{
+	if(m_buffer.get())
+		return FileManager::queryStreams(m_buffer);
+	else
+		return FileManager::queryStreams(m_filename);
+}
+
 std::shared_ptr<IReader> File::createReader()
 {
 	if(m_buffer.get())
-		return FileManager::createReader(m_buffer);
+		return FileManager::createReader(m_buffer, m_stream);
 	else
-		return FileManager::createReader(m_filename);
+		return FileManager::createReader(m_filename, m_stream);
 }
 
 AUD_NAMESPACE_END

extern/audaspace/src/file/FileManager.cpp

@@ -43,13 +43,13 @@ void FileManager::registerOutput(std::shared_ptr<aud::IFileOutput> output)
 	outputs().push_back(output);
 }
 
-std::shared_ptr<IReader> FileManager::createReader(std::string filename)
+std::shared_ptr<IReader> FileManager::createReader(std::string filename, int stream)
 {
 	for(std::shared_ptr<IFileInput> input : inputs())
 	{
 		try
 		{
-			return input->createReader(filename);
+			return input->createReader(filename, stream);
 		}
 		catch(Exception&) {}
 	}
@@ -57,13 +57,41 @@ std::shared_ptr<IReader> FileManager::createReader(std::string filename)
 	AUD_THROW(FileException, "The file couldn't be read with any installed file reader.");
 }
 
-std::shared_ptr<IReader> FileManager::createReader(std::shared_ptr<Buffer> buffer)
+std::shared_ptr<IReader> FileManager::createReader(std::shared_ptr<Buffer> buffer, int stream)
 {
 	for(std::shared_ptr<IFileInput> input : inputs())
 	{
 		try
 		{
-			return input->createReader(buffer);
+			return input->createReader(buffer, stream);
+		}
+		catch(Exception&) {}
+	}
+
+	AUD_THROW(FileException, "The file couldn't be read with any installed file reader.");
+}
+
+std::vector<StreamInfo> FileManager::queryStreams(std::string filename)
+{
+	for(std::shared_ptr<IFileInput> input : inputs())
+	{
+		try
+		{
+			return input->queryStreams(filename);
+		}
+		catch(Exception&) {}
+	}
+
+	AUD_THROW(FileException, "The file couldn't be read with any installed file reader.");
+}
+
+std::vector<StreamInfo> FileManager::queryStreams(std::shared_ptr<Buffer> buffer)
+{
+	for(std::shared_ptr<IFileInput> input : inputs())
+	{
+		try
+		{
+			return input->queryStreams(buffer);
 		}
 		catch(Exception&) {}
 	}

extern/audaspace/src/util/RingBuffer.cpp

@@ -0,0 +1,137 @@
+/*******************************************************************************
+ * Copyright 2009-2021 Jörg Müller
+ *
+ * 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.
+ ******************************************************************************/
+
+#include "util/RingBuffer.h"
+
+#include <algorithm>
+#include <cstring>
+#include <cstdlib>
+
+#define ALIGNMENT 32
+#define ALIGN(a) (a + ALIGNMENT - ((long long)a & (ALIGNMENT-1)))
+
+AUD_NAMESPACE_BEGIN
+
+RingBuffer::RingBuffer(int size) :
+	m_buffer(size),
+	m_read(0),
+	m_write(0)
+{
+}
+
+sample_t* RingBuffer::getBuffer() const
+{
+	return m_buffer.getBuffer();
+}
+
+int RingBuffer::getSize() const
+{
+	return m_buffer.getSize();
+}
+
+size_t RingBuffer::getReadSize() const
+{
+	size_t read = m_read;
+	size_t write = m_write;
+
+	if(read > write)
+		return write + getSize() - read;
+	else
+		return write - read;
+}
+
+size_t RingBuffer::getWriteSize() const
+{
+	size_t read = m_read;
+	size_t write = m_write;
+
+	if(read > write)
+		return read - write - 1;
+	else
+		return read + getSize() - write - 1;
+}
+
+size_t RingBuffer::read(data_t* target, size_t size)
+{
+	size = std::min(size, getReadSize());
+
+	data_t* buffer = reinterpret_cast<data_t*>(m_buffer.getBuffer());
+
+	if(m_read + size > m_buffer.getSize())
+	{
+		size_t read_first = m_buffer.getSize() - m_read;
+		size_t read_second = size - read_first;
+
+		std::memcpy(target, buffer + m_read, read_first);
+		std::memcpy(target + read_first, buffer, read_second);
+
+		m_read = read_second;
+	}
+	else
+	{
+		std::memcpy(target, buffer + m_read, size);
+
+		m_read += size;
+	}
+
+	return size;
+}
+
+size_t RingBuffer::write(data_t* source, size_t size)
+{
+	size = std::min(size, getWriteSize());
+
+	data_t* buffer = reinterpret_cast<data_t*>(m_buffer.getBuffer());
+
+	if(m_write + size > m_buffer.getSize())
+	{
+		size_t write_first = m_buffer.getSize() - m_write;
+		size_t write_second = size - write_first;
+
+		std::memcpy(buffer + m_write, source, write_first);
+		std::memcpy(buffer, source + write_first, write_second);
+
+		m_write = write_second;
+	}
+	else
+	{
+		std::memcpy(buffer + m_write, source, size);
+
+		m_write += size;
+	}
+
+	return size;
+}
+
+void RingBuffer::reset()
+{
+	m_read = 0;
+	m_write = 0;
+}
+
+void RingBuffer::resize(int size)
+{
+	m_buffer.resize(size);
+	reset();
+}
+
+void RingBuffer::assureSize(int size)
+{
+	m_buffer.assureSize(size);
+	reset();
+}
+
+AUD_NAMESPACE_END

extern/cuew/include/cuew.h

@@ -645,7 +645,8 @@ typedef enum CUdevice_P2PAttribute_enum {
   CU_DEVICE_P2P_ATTRIBUTE_PERFORMANCE_RANK = 0x01,
   CU_DEVICE_P2P_ATTRIBUTE_ACCESS_SUPPORTED = 0x02,
   CU_DEVICE_P2P_ATTRIBUTE_NATIVE_ATOMIC_SUPPORTED = 0x03,
-  CU_DEVICE_P2P_ATTRIBUTE_ARRAY_ACCESS_ACCESS_SUPPORTED = 0x04,
+  CU_DEVICE_P2P_ATTRIBUTE_ACCESS_ACCESS_SUPPORTED = 0x04,
+  CU_DEVICE_P2P_ATTRIBUTE_CUDA_ARRAY_ACCESS_SUPPORTED = 0x04,
 } CUdevice_P2PAttribute;
 
 typedef void (CUDA_CB *CUstreamCallback)(CUstream hStream, CUresult status, void* userData);

extern/hipew/CMakeLists.txt

@@ -0,0 +1,39 @@
+# ***** BEGIN GPL LICENSE BLOCK *****
+#
+# This program is free software; you can redistribute it and/or
+# modify it under the terms of the GNU General Public License
+# as published by the Free Software Foundation; either version 2
+# of the License, or (at your option) any later version.
+#
+# This program is distributed in the hope that it will be useful,
+# but WITHOUT ANY WARRANTY; without even the implied warranty of
+# MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the
+# GNU General Public License for more details.
+#
+# You should have received a copy of the GNU General Public License
+# along with this program; if not, write to the Free Software Foundation,
+# Inc., 51 Franklin Street, Fifth Floor, Boston, MA 02110-1301, USA.
+#
+# The Original Code is Copyright (C) 2021, Blender Foundation
+# All rights reserved.
+# ***** END GPL LICENSE BLOCK *****
+
+set(INC
+  .
+  include
+)
+
+set(INC_SYS
+
+)
+
+set(SRC
+  src/hipew.c
+
+  include/hipew.h
+)
+
+set(LIB
+)
+
+blender_add_lib(extern_hipew "${SRC}" "${INC}" "${INC_SYS}" "${LIB}")

extern/hipew/src/hipew.c

@@ -0,0 +1,533 @@
+/*
+ * Copyright 2011-2021 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
+ */
+#ifdef _MSC_VER
+#  if _MSC_VER < 1900
+#    define snprintf _snprintf
+#  endif
+#  define popen _popen
+#  define pclose _pclose
+#  define _CRT_SECURE_NO_WARNINGS
+#endif
+
+#include <hipew.h>
+#include <assert.h>
+#include <stdio.h>
+#include <string.h>
+#include <sys/stat.h>
+
+#ifdef _WIN32
+#  define WIN32_LEAN_AND_MEAN
+#  define VC_EXTRALEAN
+#  include <windows.h>
+
+/* Utility macros. */
+
+typedef HMODULE DynamicLibrary;
+
+#  define dynamic_library_open(path)         LoadLibraryA(path)
+#  define dynamic_library_close(lib)         FreeLibrary(lib)
+#  define dynamic_library_find(lib, symbol)  GetProcAddress(lib, symbol)
+#else
+#  include <dlfcn.h>
+
+typedef void* DynamicLibrary;
+
+#  define dynamic_library_open(path)         dlopen(path, RTLD_NOW)
+#  define dynamic_library_close(lib)         dlclose(lib)
+#  define dynamic_library_find(lib, symbol)  dlsym(lib, symbol)
+#endif
+
+#define _LIBRARY_FIND_CHECKED(lib, name) \
+        name = (t##name *)dynamic_library_find(lib, #name); \
+        assert(name);
+
+#define _LIBRARY_FIND(lib, name) \
+        name = (t##name *)dynamic_library_find(lib, #name);
+
+#define HIP_LIBRARY_FIND_CHECKED(name) \
+        _LIBRARY_FIND_CHECKED(hip_lib, name)
+#define HIP_LIBRARY_FIND(name) _LIBRARY_FIND(hip_lib, name)
+
+
+static DynamicLibrary hip_lib;
+
+/* Function definitions. */
+thipGetErrorName *hipGetErrorName;
+thipInit *hipInit;
+thipDriverGetVersion *hipDriverGetVersion;
+thipGetDevice *hipGetDevice;
+thipGetDeviceCount *hipGetDeviceCount;
+thipDeviceGetName *hipDeviceGetName;
+thipDeviceGetAttribute *hipDeviceGetAttribute;
+thipDeviceComputeCapability *hipDeviceComputeCapability;
+thipDevicePrimaryCtxRetain *hipDevicePrimaryCtxRetain;
+thipDevicePrimaryCtxRelease *hipDevicePrimaryCtxRelease;
+thipDevicePrimaryCtxSetFlags *hipDevicePrimaryCtxSetFlags;
+thipDevicePrimaryCtxGetState *hipDevicePrimaryCtxGetState;
+thipDevicePrimaryCtxReset *hipDevicePrimaryCtxReset;
+thipCtxCreate *hipCtxCreate;
+thipCtxDestroy *hipCtxDestroy;
+thipCtxPushCurrent *hipCtxPushCurrent;
+thipCtxPopCurrent *hipCtxPopCurrent;
+thipCtxSetCurrent *hipCtxSetCurrent;
+thipCtxGetCurrent *hipCtxGetCurrent;
+thipCtxGetDevice *hipCtxGetDevice;
+thipCtxGetFlags *hipCtxGetFlags;
+thipCtxSynchronize *hipCtxSynchronize;
+thipDeviceSynchronize *hipDeviceSynchronize;
+thipCtxGetCacheConfig *hipCtxGetCacheConfig;
+thipCtxSetCacheConfig *hipCtxSetCacheConfig;
+thipCtxGetSharedMemConfig *hipCtxGetSharedMemConfig;
+thipCtxSetSharedMemConfig *hipCtxSetSharedMemConfig;
+thipCtxGetApiVersion *hipCtxGetApiVersion;
+thipModuleLoad *hipModuleLoad;
+thipModuleLoadData *hipModuleLoadData;
+thipModuleLoadDataEx *hipModuleLoadDataEx;
+thipModuleUnload *hipModuleUnload;
+thipModuleGetFunction *hipModuleGetFunction;
+thipModuleGetGlobal *hipModuleGetGlobal;
+thipModuleGetTexRef *hipModuleGetTexRef;
+thipMemGetInfo *hipMemGetInfo;
+thipMalloc *hipMalloc;
+thipMemAllocPitch *hipMemAllocPitch;
+thipFree *hipFree;
+thipMemGetAddressRange *hipMemGetAddressRange;
+thipHostMalloc *hipHostMalloc;
+thipHostFree *hipHostFree;
+thipHostGetDevicePointer *hipHostGetDevicePointer;
+thipHostGetFlags *hipHostGetFlags;
+thipMallocManaged *hipMallocManaged;
+thipDeviceGetByPCIBusId *hipDeviceGetByPCIBusId;
+thipDeviceGetPCIBusId *hipDeviceGetPCIBusId;
+thipMemcpyPeer *hipMemcpyPeer;
+thipMemcpyHtoD *hipMemcpyHtoD;
+thipMemcpyDtoH *hipMemcpyDtoH;
+thipMemcpyDtoD *hipMemcpyDtoD;
+thipDrvMemcpy2DUnaligned *hipDrvMemcpy2DUnaligned;
+thipMemcpyParam2D *hipMemcpyParam2D;
+thipDrvMemcpy3D *hipDrvMemcpy3D;
+thipMemcpyHtoDAsync *hipMemcpyHtoDAsync;
+thipMemcpyDtoHAsync *hipMemcpyDtoHAsync;
+thipMemcpyParam2DAsync *hipMemcpyParam2DAsync;
+thipDrvMemcpy3DAsync *hipDrvMemcpy3DAsync;
+thipMemsetD8 *hipMemsetD8;
+thipMemsetD16 *hipMemsetD16;
+thipMemsetD32 *hipMemsetD32;
+thipMemsetD8Async *hipMemsetD8Async;
+thipMemsetD16Async *hipMemsetD16Async;
+thipMemsetD32Async *hipMemsetD32Async;
+thipArrayCreate *hipArrayCreate;
+thipArrayDestroy *hipArrayDestroy;
+thipArray3DCreate *hipArray3DCreate;
+thipStreamCreateWithFlags *hipStreamCreateWithFlags;
+thipStreamCreateWithPriority *hipStreamCreateWithPriority;
+thipStreamGetPriority *hipStreamGetPriority;
+thipStreamGetFlags *hipStreamGetFlags;
+thipStreamWaitEvent *hipStreamWaitEvent;
+thipStreamAddCallback *hipStreamAddCallback;
+thipStreamQuery *hipStreamQuery;
+thipStreamSynchronize *hipStreamSynchronize;
+thipStreamDestroy *hipStreamDestroy;
+thipEventCreateWithFlags *hipEventCreateWithFlags;
+thipEventRecord *hipEventRecord;
+thipEventQuery *hipEventQuery;
+thipEventSynchronize *hipEventSynchronize;
+thipEventDestroy *hipEventDestroy;
+thipEventElapsedTime *hipEventElapsedTime;
+thipFuncGetAttribute *hipFuncGetAttribute;
+thipFuncSetCacheConfig *hipFuncSetCacheConfig;
+thipModuleLaunchKernel *hipModuleLaunchKernel;
+thipDrvOccupancyMaxActiveBlocksPerMultiprocessor *hipDrvOccupancyMaxActiveBlocksPerMultiprocessor;
+thipDrvOccupancyMaxActiveBlocksPerMultiprocessorWithFlags *hipDrvOccupancyMaxActiveBlocksPerMultiprocessorWithFlags;
+thipModuleOccupancyMaxPotentialBlockSize *hipModuleOccupancyMaxPotentialBlockSize;
+thipTexRefSetArray *hipTexRefSetArray;
+thipTexRefSetAddress *hipTexRefSetAddress;
+thipTexRefSetAddress2D *hipTexRefSetAddress2D;
+thipTexRefSetFormat *hipTexRefSetFormat;
+thipTexRefSetAddressMode *hipTexRefSetAddressMode;
+thipTexRefSetFilterMode *hipTexRefSetFilterMode;
+thipTexRefSetFlags *hipTexRefSetFlags;
+thipTexRefGetAddress *hipTexRefGetAddress;
+thipTexRefGetArray *hipTexRefGetArray;
+thipTexRefGetAddressMode *hipTexRefGetAddressMode;
+thipTexObjectCreate *hipTexObjectCreate;
+thipTexObjectDestroy *hipTexObjectDestroy;
+thipDeviceCanAccessPeer *hipDeviceCanAccessPeer;
+
+thipCtxEnablePeerAccess *hipCtxEnablePeerAccess;
+thipCtxDisablePeerAccess *hipCtxDisablePeerAccess;
+thipDeviceGetP2PAttribute *hipDeviceGetP2PAttribute;
+thipGraphicsUnregisterResource *hipGraphicsUnregisterResource;
+thipGraphicsMapResources *hipGraphicsMapResources;
+thipGraphicsUnmapResources *hipGraphicsUnmapResources;
+thipGraphicsResourceGetMappedPointer *hipGraphicsResourceGetMappedPointer;
+
+thipGraphicsGLRegisterBuffer *hipGraphicsGLRegisterBuffer;
+thipGLGetDevices *hipGLGetDevices;
+
+
+
+static DynamicLibrary dynamic_library_open_find(const char **paths) {
+  int i = 0;
+  while (paths[i] != NULL) {
+      DynamicLibrary lib = dynamic_library_open(paths[i]);
+      if (lib != NULL) {
+        return lib;
+      }
+      ++i;
+  }
+  return NULL;
+}
+
+/* Implementation function. */
+static void hipewHipExit(void) {
+  if (hip_lib != NULL) {
+    /*  Ignore errors. */
+    dynamic_library_close(hip_lib);
+    hip_lib = NULL;
+  }
+}
+
+static int hipewHipInit(void) {
+  /* Library paths. */
+#ifdef _WIN32
+  /* Expected in c:/windows/system or similar, no path needed. */
+  const char *hip_paths[] = {"amdhip64.dll", NULL};
+#elif defined(__APPLE__)
+  /* Default installation path. */
+  const char *hip_paths[] = {"", NULL};
+#else
+  const char *hip_paths[] = {"/opt/rocm/hip/lib/libamdhip64.so", NULL};
+#endif
+  static int initialized = 0;
+  static int result = 0;
+  int error, driver_version;
+
+  if (initialized) {
+    return result;
+  }
+
+  initialized = 1;
+
+  error = atexit(hipewHipExit);
+  if (error) {
+    result = HIPEW_ERROR_ATEXIT_FAILED;
+    return result;
+  }
+
+  /* Load library. */
+  hip_lib = dynamic_library_open_find(hip_paths);
+
+  if (hip_lib == NULL) {
+    result = HIPEW_ERROR_OPEN_FAILED;
+    return result;
+  }
+
+  /* Fetch all function pointers. */
+  HIP_LIBRARY_FIND_CHECKED(hipGetErrorName);
+  HIP_LIBRARY_FIND_CHECKED(hipInit);
+  HIP_LIBRARY_FIND_CHECKED(hipDriverGetVersion);
+  HIP_LIBRARY_FIND_CHECKED(hipGetDevice);
+  HIP_LIBRARY_FIND_CHECKED(hipGetDeviceCount);
+  HIP_LIBRARY_FIND_CHECKED(hipDeviceGetName);
+  HIP_LIBRARY_FIND_CHECKED(hipDeviceGetAttribute);
+  HIP_LIBRARY_FIND_CHECKED(hipDeviceComputeCapability);
+  HIP_LIBRARY_FIND_CHECKED(hipDevicePrimaryCtxRetain);
+  HIP_LIBRARY_FIND_CHECKED(hipDevicePrimaryCtxRelease);
+  HIP_LIBRARY_FIND_CHECKED(hipDevicePrimaryCtxSetFlags);
+  HIP_LIBRARY_FIND_CHECKED(hipDevicePrimaryCtxGetState);
+  HIP_LIBRARY_FIND_CHECKED(hipDevicePrimaryCtxReset);
+  HIP_LIBRARY_FIND_CHECKED(hipCtxCreate);
+  HIP_LIBRARY_FIND_CHECKED(hipCtxDestroy);
+  HIP_LIBRARY_FIND_CHECKED(hipCtxPushCurrent);
+  HIP_LIBRARY_FIND_CHECKED(hipCtxPopCurrent);
+  HIP_LIBRARY_FIND_CHECKED(hipCtxSetCurrent);
+  HIP_LIBRARY_FIND_CHECKED(hipCtxGetCurrent);
+  HIP_LIBRARY_FIND_CHECKED(hipCtxGetDevice);
+  HIP_LIBRARY_FIND_CHECKED(hipCtxGetFlags);
+  HIP_LIBRARY_FIND_CHECKED(hipCtxSynchronize);
+  HIP_LIBRARY_FIND_CHECKED(hipDeviceSynchronize);
+  HIP_LIBRARY_FIND_CHECKED(hipCtxGetCacheConfig);
+  HIP_LIBRARY_FIND_CHECKED(hipCtxSetCacheConfig);
+  HIP_LIBRARY_FIND_CHECKED(hipCtxGetSharedMemConfig);
+  HIP_LIBRARY_FIND_CHECKED(hipCtxSetSharedMemConfig);
+  HIP_LIBRARY_FIND_CHECKED(hipCtxGetApiVersion);
+  HIP_LIBRARY_FIND_CHECKED(hipModuleLoad);
+  HIP_LIBRARY_FIND_CHECKED(hipModuleLoadData);
+  HIP_LIBRARY_FIND_CHECKED(hipModuleLoadDataEx);
+  HIP_LIBRARY_FIND_CHECKED(hipModuleUnload);
+  HIP_LIBRARY_FIND_CHECKED(hipModuleGetFunction);
+  HIP_LIBRARY_FIND_CHECKED(hipModuleGetGlobal);
+  HIP_LIBRARY_FIND_CHECKED(hipModuleGetTexRef);
+  HIP_LIBRARY_FIND_CHECKED(hipMemGetInfo);
+  HIP_LIBRARY_FIND_CHECKED(hipMalloc);
+  HIP_LIBRARY_FIND_CHECKED(hipMemAllocPitch);
+  HIP_LIBRARY_FIND_CHECKED(hipFree);
+  HIP_LIBRARY_FIND_CHECKED(hipMemGetAddressRange);
+  HIP_LIBRARY_FIND_CHECKED(hipHostMalloc);
+  HIP_LIBRARY_FIND_CHECKED(hipHostFree);
+  HIP_LIBRARY_FIND_CHECKED(hipHostGetDevicePointer);
+  HIP_LIBRARY_FIND_CHECKED(hipHostGetFlags);
+  HIP_LIBRARY_FIND_CHECKED(hipMallocManaged);
+  HIP_LIBRARY_FIND_CHECKED(hipDeviceGetByPCIBusId);
+  HIP_LIBRARY_FIND_CHECKED(hipDeviceGetPCIBusId);
+  HIP_LIBRARY_FIND_CHECKED(hipMemcpyPeer);
+  HIP_LIBRARY_FIND_CHECKED(hipMemcpyHtoD);
+  HIP_LIBRARY_FIND_CHECKED(hipMemcpyDtoH);
+  HIP_LIBRARY_FIND_CHECKED(hipMemcpyDtoD);
+  HIP_LIBRARY_FIND_CHECKED(hipMemcpyParam2D);
+  HIP_LIBRARY_FIND_CHECKED(hipDrvMemcpy3D);
+  HIP_LIBRARY_FIND_CHECKED(hipMemcpyHtoDAsync);
+  HIP_LIBRARY_FIND_CHECKED(hipMemcpyDtoHAsync);
+  HIP_LIBRARY_FIND_CHECKED(hipDrvMemcpy2DUnaligned);
+  HIP_LIBRARY_FIND_CHECKED(hipMemcpyParam2DAsync);
+  HIP_LIBRARY_FIND_CHECKED(hipDrvMemcpy3DAsync);
+  HIP_LIBRARY_FIND_CHECKED(hipMemsetD8);
+  HIP_LIBRARY_FIND_CHECKED(hipMemsetD16);
+  HIP_LIBRARY_FIND_CHECKED(hipMemsetD32);
+  HIP_LIBRARY_FIND_CHECKED(hipMemsetD8Async);
+  HIP_LIBRARY_FIND_CHECKED(hipMemsetD16Async);
+  HIP_LIBRARY_FIND_CHECKED(hipMemsetD32Async);
+  HIP_LIBRARY_FIND_CHECKED(hipArrayCreate);
+  HIP_LIBRARY_FIND_CHECKED(hipArrayDestroy);
+  HIP_LIBRARY_FIND_CHECKED(hipArray3DCreate);
+  HIP_LIBRARY_FIND_CHECKED(hipStreamCreateWithFlags);
+  HIP_LIBRARY_FIND_CHECKED(hipStreamCreateWithPriority);
+  HIP_LIBRARY_FIND_CHECKED(hipStreamGetPriority);
+  HIP_LIBRARY_FIND_CHECKED(hipStreamGetFlags);
+  HIP_LIBRARY_FIND_CHECKED(hipStreamWaitEvent);
+  HIP_LIBRARY_FIND_CHECKED(hipStreamAddCallback);
+  HIP_LIBRARY_FIND_CHECKED(hipStreamQuery);
+  HIP_LIBRARY_FIND_CHECKED(hipStreamSynchronize);
+  HIP_LIBRARY_FIND_CHECKED(hipStreamDestroy);
+  HIP_LIBRARY_FIND_CHECKED(hipEventCreateWithFlags);
+  HIP_LIBRARY_FIND_CHECKED(hipEventRecord);
+  HIP_LIBRARY_FIND_CHECKED(hipEventQuery);
+  HIP_LIBRARY_FIND_CHECKED(hipEventSynchronize);
+  HIP_LIBRARY_FIND_CHECKED(hipEventDestroy);
+  HIP_LIBRARY_FIND_CHECKED(hipEventElapsedTime);
+  HIP_LIBRARY_FIND_CHECKED(hipFuncGetAttribute);
+  HIP_LIBRARY_FIND_CHECKED(hipFuncSetCacheConfig);
+  HIP_LIBRARY_FIND_CHECKED(hipModuleLaunchKernel);
+  HIP_LIBRARY_FIND_CHECKED(hipModuleOccupancyMaxPotentialBlockSize);
+  HIP_LIBRARY_FIND_CHECKED(hipTexRefSetArray);
+  HIP_LIBRARY_FIND_CHECKED(hipTexRefSetAddress);
+  HIP_LIBRARY_FIND_CHECKED(hipTexRefSetAddress2D);
+  HIP_LIBRARY_FIND_CHECKED(hipTexRefSetFormat);
+  HIP_LIBRARY_FIND_CHECKED(hipTexRefSetAddressMode);
+  HIP_LIBRARY_FIND_CHECKED(hipTexRefSetFilterMode);
+  HIP_LIBRARY_FIND_CHECKED(hipTexRefSetFlags);
+  HIP_LIBRARY_FIND_CHECKED(hipTexRefGetAddress);
+  HIP_LIBRARY_FIND_CHECKED(hipTexRefGetAddressMode);
+  HIP_LIBRARY_FIND_CHECKED(hipTexObjectCreate);
+  HIP_LIBRARY_FIND_CHECKED(hipTexObjectDestroy);
+  HIP_LIBRARY_FIND_CHECKED(hipDeviceCanAccessPeer);
+  HIP_LIBRARY_FIND_CHECKED(hipCtxEnablePeerAccess);
+  HIP_LIBRARY_FIND_CHECKED(hipCtxDisablePeerAccess);
+  HIP_LIBRARY_FIND_CHECKED(hipDeviceGetP2PAttribute);
+#ifdef _WIN32
+  HIP_LIBRARY_FIND_CHECKED(hipGraphicsUnregisterResource);
+  HIP_LIBRARY_FIND_CHECKED(hipGraphicsMapResources);
+  HIP_LIBRARY_FIND_CHECKED(hipGraphicsUnmapResources);
+  HIP_LIBRARY_FIND_CHECKED(hipGraphicsResourceGetMappedPointer);
+  HIP_LIBRARY_FIND_CHECKED(hipGraphicsGLRegisterBuffer);
+  HIP_LIBRARY_FIND_CHECKED(hipGLGetDevices);
+#endif
+  result = HIPEW_SUCCESS;
+  return result;
+}
+
+
+
+int hipewInit(hipuint32_t flags) {
+  int result = HIPEW_SUCCESS;
+
+  if (flags & HIPEW_INIT_HIP) {
+    result = hipewHipInit();
+    if (result != HIPEW_SUCCESS) {
+      return result;
+    }
+  }
+
+  return result;
+}
+
+
+const char *hipewErrorString(hipError_t result) {
+  switch (result) {
+    case hipSuccess: return "No errors";
+    case hipErrorInvalidValue: return "Invalid value";
+    case hipErrorOutOfMemory: return "Out of memory";
+    case hipErrorNotInitialized: return "Driver not initialized";
+    case hipErrorDeinitialized: return "Driver deinitialized";
+    case hipErrorProfilerDisabled: return "Profiler disabled";
+    case hipErrorProfilerNotInitialized: return "Profiler not initialized";
+    case hipErrorProfilerAlreadyStarted: return "Profiler already started";
+    case hipErrorProfilerAlreadyStopped: return "Profiler already stopped";
+    case hipErrorNoDevice: return "No HIP-capable device available";
+    case hipErrorInvalidDevice: return "Invalid device";
+    case hipErrorInvalidImage: return "Invalid kernel image";
+    case hipErrorInvalidContext: return "Invalid context";
+    case hipErrorContextAlreadyCurrent: return "Context already current";
+    case hipErrorMapFailed: return "Map failed";
+    case hipErrorUnmapFailed: return "Unmap failed";
+    case hipErrorArrayIsMapped: return "Array is mapped";
+    case hipErrorAlreadyMapped: return "Already mapped";
+    case hipErrorNoBinaryForGpu: return "No binary for GPU";
+    case hipErrorAlreadyAcquired: return "Already acquired";
+    case hipErrorNotMapped: return "Not mapped";
+    case hipErrorNotMappedAsArray: return "Mapped resource not available for access as an array";
+    case hipErrorNotMappedAsPointer: return "Mapped resource not available for access as a pointer";
+    case hipErrorECCNotCorrectable: return "Uncorrectable ECC error detected";
+    case hipErrorUnsupportedLimit: return "hipLimit_t not supported by device";
+    case hipErrorContextAlreadyInUse: return "Context already in use";
+    case hipErrorPeerAccessUnsupported: return "Peer access unsupported";
+    case hipErrorInvalidKernelFile: return "Invalid ptx";
+    case hipErrorInvalidGraphicsContext: return "Invalid graphics context";
+    case hipErrorInvalidSource: return "Invalid source";
+    case hipErrorFileNotFound: return "File not found";
+    case hipErrorSharedObjectSymbolNotFound: return "Link to a shared object failed to resolve";
+    case hipErrorSharedObjectInitFailed: return "Shared object initialization failed";
+    case hipErrorOperatingSystem: return "Operating system";
+    case hipErrorInvalidHandle: return "Invalid handle";
+    case hipErrorNotFound: return "Not found";
+    case hipErrorNotReady: return "HIP not ready";
+    case hipErrorIllegalAddress: return "Illegal address";
+    case hipErrorLaunchOutOfResources: return "Launch exceeded resources";
+    case hipErrorLaunchTimeOut: return "Launch exceeded timeout";
+    case hipErrorPeerAccessAlreadyEnabled: return "Peer access already enabled";
+    case hipErrorPeerAccessNotEnabled: return "Peer access not enabled";
+    case hipErrorSetOnActiveProcess: return "Primary context active";
+    case hipErrorAssert: return "Assert";
+    case hipErrorHostMemoryAlreadyRegistered: return "Host memory already registered";
+    case hipErrorHostMemoryNotRegistered: return "Host memory not registered";
+    case hipErrorLaunchFailure: return "Launch failed";
+    case hipErrorCooperativeLaunchTooLarge: return "Cooperative launch too large";
+    case hipErrorNotSupported: return "Not supported";
+    case hipErrorUnknown: return "Unknown error";
+    default: return "Unknown HIP error value";
+  }
+}
+
+static void path_join(const char *path1,
+                      const char *path2,
+                      int maxlen,
+                      char *result) {
+#if defined(WIN32) || defined(_WIN32)
+  const char separator = '\\';
+#else
+  const char separator = '/';
+#endif
+  int n = snprintf(result, maxlen, "%s%c%s", path1, separator, path2);
+  if (n != -1 && n < maxlen) {
+    result[n] = '\0';
+  }
+  else {
+    result[maxlen - 1] = '\0';
+  }
+}
+
+static int path_exists(const char *path) {
+  struct stat st;
+  if (stat(path, &st)) {
+    return 0;
+  }
+  return 1;
+}
+
+const char *hipewCompilerPath(void) {
+    #ifdef _WIN32
+    const char *hipPath = getenv("HIP_ROCCLR_HOME");
+    const char *windowsCommand = "perl ";
+    const char *executable = "bin/hipcc";
+
+    static char hipcc[65536];
+    static char finalCommand[65536];
+    if(hipPath) {
+      path_join(hipPath, executable, sizeof(hipcc), hipcc);
+      if(path_exists(hipcc)) {
+        snprintf(finalCommand, sizeof(hipcc), "%s %s", windowsCommand, hipcc);
+        return finalCommand;
+      } else {
+        printf("Could not find hipcc. Make sure HIP_ROCCLR_HOME points to the directory holding /bin/hipcc");
+      }
+    }
+    #else
+    const char *hipPath =  "opt/rocm/hip/bin";
+    const char *executable = "hipcc";
+
+    static char hipcc[65536];
+    if(hipPath) {
+      path_join(hipPath, executable, sizeof(hipcc), hipcc);
+      if(path_exists(hipcc)){
+        return hipcc;
+      }
+    }
+    #endif
+
+  {
+#ifdef _WIN32
+    FILE *handle = popen("where hipcc", "r");
+#else
+    FILE *handle = popen("which hipcc", "r");
+#endif
+    if (handle) {
+      char buffer[4096] = {0};
+      int len = fread(buffer, 1, sizeof(buffer) - 1, handle);
+      buffer[len] = '\0';
+      pclose(handle);
+      if (buffer[0]) {
+        return "hipcc";
+      }
+    }
+  }
+
+  return NULL;
+}
+
+int hipewCompilerVersion(void) {
+  const char *path = hipewCompilerPath();
+  const char *marker = "Hip compilation tools, release ";
+  FILE *pipe;
+  int major, minor;
+  char *versionstr;
+  char buf[128];
+  char output[65536] = "\0";
+  char command[65536] = "\0";
+
+  if (path == NULL) {
+    return 0;
+  }
+
+  /* get --version output */
+  strcat(command, "\"");
+  strncat(command, path, sizeof(command) - 1);
+  strncat(command, "\" --version", sizeof(command) - strlen(path) - 1);
+  pipe = popen(command, "r");
+  if (!pipe) {
+    fprintf(stderr, "HIP: failed to run compiler to retrieve version");
+    return 0;
+  }
+
+  while (!feof(pipe)) {
+    if (fgets(buf, sizeof(buf), pipe) != NULL) {
+      strncat(output, buf, sizeof(output) - strlen(output) - 1);
+    }
+  }
+
+  pclose(pipe);
+  return 40;
+}

extern/json/README.blender

@@ -0,0 +1,5 @@
+Project: JSON
+URL: https://github.com/nlohmann/json/
+License: MIT License
+Upstream version: 3.10.2
+Local modifications: None

extern/tinygltf/README.blender

@@ -0,0 +1,6 @@
+Project: TinyGLTF
+URL: https://github.com/syoyo/tinygltf
+License: MIT
+Upstream version: 2.5.0, 19a41d20ec0
+Local modifications: 
+* Silence "enum value not handled in switch" warnings due to JSON dependency.

intern/cycles/CMakeLists.txt

@@ -247,7 +247,7 @@ if(WITH_CYCLES_OSL)
 endif()
 
 if(WITH_CYCLES_DEVICE_OPTIX)
-  find_package(OptiX)
+  find_package(OptiX 7.3.0)
 
   if(OPTIX_FOUND)
     add_definitions(-DWITH_OPTIX)
@@ -286,11 +286,18 @@ if(WITH_OPENSUBDIV)
   )
 endif()
 
+if(WITH_OPENIMAGEDENOISE)
+  add_definitions(-DWITH_OPENIMAGEDENOISE)
+  add_definitions(-DOIDN_STATIC_LIB)
+  include_directories(
+    SYSTEM
+    ${OPENIMAGEDENOISE_INCLUDE_DIRS}
+  )
+endif()
+
 if(WITH_CYCLES_STANDALONE)
-  set(WITH_CYCLES_DEVICE_OPENCL TRUE)
   set(WITH_CYCLES_DEVICE_CUDA TRUE)
-  # Experimental and unfinished.
-  set(WITH_CYCLES_NETWORK FALSE)
+  set(WITH_CYCLES_DEVICE_HIP TRUE)
 endif()
 # TODO(sergey): Consider removing it, only causes confusion in interface.
 set(WITH_CYCLES_DEVICE_MULTI TRUE)
@@ -386,18 +393,12 @@ if(WITH_CYCLES_BLENDER)
   add_subdirectory(blender)
 endif()
 
-if(WITH_CYCLES_NETWORK)
-  add_definitions(-DWITH_NETWORK)
-endif()
-
-if(WITH_CYCLES_STANDALONE OR WITH_CYCLES_NETWORK OR WITH_CYCLES_CUBIN_COMPILER)
-  add_subdirectory(app)
-endif()
-
+add_subdirectory(app)
 add_subdirectory(bvh)
 add_subdirectory(device)
 add_subdirectory(doc)
 add_subdirectory(graph)
+add_subdirectory(integrator)
 add_subdirectory(kernel)
 add_subdirectory(render)
 add_subdirectory(subd)

intern/cycles/app/CMakeLists.txt

@@ -90,24 +90,6 @@ if(WITH_CYCLES_STANDALONE)
   endif()
 endif()
 
-#####################################################################
-# Cycles network server executable
-#####################################################################
-
-if(WITH_CYCLES_NETWORK)
-  set(SRC
-    cycles_server.cpp
-  )
-  add_executable(cycles_server ${SRC})
-  target_link_libraries(cycles_server ${LIBRARIES})
-  cycles_target_link_libraries(cycles_server)
-
-  if(UNIX AND NOT APPLE)
-    set_target_properties(cycles_server PROPERTIES INSTALL_RPATH $ORIGIN/lib)
-  endif()
-  unset(SRC)
-endif()
-
 #####################################################################
 # Cycles cubin compiler executable
 #####################################################################

intern/cycles/app/cycles_standalone.cpp

@@ -53,7 +53,7 @@ struct Options {
   SessionParams session_params;
   bool quiet;
   bool show_help, interactive, pause;
-  string output_path;
+  string output_filepath;
 } options;
 
 static void session_print(const string &str)
@@ -126,7 +126,7 @@ static BufferParams &session_buffer_params()
 
 static void scene_init()
 {
-  options.scene = new Scene(options.scene_params, options.session->device);
+  options.scene = options.session->scene;
 
   /* Read XML */
   xml_read_file(options.scene, options.filepath.c_str());
@@ -148,7 +148,7 @@ static void scene_init()
 static void session_init()
 {
   options.session_params.write_render_cb = write_render;
-  options.session = new Session(options.session_params);
+  options.session = new Session(options.session_params, options.scene_params);
 
   if (options.session_params.background && !options.quiet)
     options.session->progress.set_update_callback(function_bind(&session_print_status));
@@ -159,9 +159,8 @@ static void session_init()
 
   /* load scene */
   scene_init();
-  options.session->scene = options.scene;
 
-  options.session->reset(session_buffer_params(), options.session_params.samples);
+  options.session->reset(options.session_params, session_buffer_params());
   options.session->start();
 }
 
@@ -223,9 +222,7 @@ static void display_info(Progress &progress)
 
 static void display()
 {
-  static DeviceDrawParams draw_params = DeviceDrawParams();
-
-  options.session->draw(session_buffer_params(), draw_params);
+  options.session->draw();
 
   display_info(options.session->progress);
 }
@@ -255,7 +252,7 @@ static void motion(int x, int y, int button)
     options.session->scene->camera->need_flags_update = true;
     options.session->scene->camera->need_device_update = true;
 
-    options.session->reset(session_buffer_params(), options.session_params.samples);
+    options.session->reset(options.session_params, session_buffer_params());
   }
 }
 
@@ -272,7 +269,7 @@ static void resize(int width, int height)
     options.session->scene->camera->need_flags_update = true;
     options.session->scene->camera->need_device_update = true;
 
-    options.session->reset(session_buffer_params(), options.session_params.samples);
+    options.session->reset(options.session_params, session_buffer_params());
   }
 }
 
@@ -284,7 +281,7 @@ static void keyboard(unsigned char key)
 
   /* Reset */
   else if (key == 'r')
-    options.session->reset(session_buffer_params(), options.session_params.samples);
+    options.session->reset(options.session_params, session_buffer_params());
 
   /* Cancel */
   else if (key == 27)  // escape
@@ -321,7 +318,7 @@ static void keyboard(unsigned char key)
     options.session->scene->camera->need_flags_update = true;
     options.session->scene->camera->need_device_update = true;
 
-    options.session->reset(session_buffer_params(), options.session_params.samples);
+    options.session->reset(options.session_params, session_buffer_params());
   }
 
   /* Set Max Bounces */
@@ -347,7 +344,7 @@ static void keyboard(unsigned char key)
 
     options.session->scene->integrator->set_max_bounce(bounce);
 
-    options.session->reset(session_buffer_params(), options.session_params.samples);
+    options.session->reset(options.session_params, session_buffer_params());
   }
 }
 #endif
@@ -362,11 +359,13 @@ static int files_parse(int argc, const char *argv[])
 
 static void options_parse(int argc, const char **argv)
 {
-  options.width = 0;
-  options.height = 0;
+  options.width = 1024;
+  options.height = 512;
   options.filepath = "";
   options.session = NULL;
   options.quiet = false;
+  options.session_params.use_auto_tile = false;
+  options.session_params.tile_size = 0;
 
   /* device names */
   string device_names = "";
@@ -412,7 +411,7 @@ static void options_parse(int argc, const char **argv)
              &options.session_params.samples,
              "Number of samples to render",
              "--output %s",
-             &options.output_path,
+             &options.output_filepath,
              "File path to write output image",
              "--threads %d",
              &options.session_params.threads,
@@ -423,12 +422,9 @@ static void options_parse(int argc, const char **argv)
              "--height %d",
              &options.height,
              "Window height in pixel",
-             "--tile-width %d",
-             &options.session_params.tile_size.x,
-             "Tile width in pixels",
-             "--tile-height %d",
-             &options.session_params.tile_size.y,
-             "Tile height in pixels",
+             "--tile-size %d",
+             &options.session_params.tile_size,
+             "Tile size in pixels",
              "--list-devices",
              &list,
              "List information about all available devices",
@@ -490,8 +486,9 @@ static void options_parse(int argc, const char **argv)
   options.session_params.background = true;
 #endif
 
-  /* Use progressive rendering */
-  options.session_params.progressive = true;
+  if (options.session_params.tile_size > 0) {
+    options.session_params.use_auto_tile = true;
+  }
 
   /* find matching device */
   DeviceType device_type = Device::type_from_string(devicename.c_str());
@@ -527,9 +524,6 @@ static void options_parse(int argc, const char **argv)
     fprintf(stderr, "No file path specified\n");
     exit(EXIT_FAILURE);
   }
-
-  /* For smoother Viewport */
-  options.session_params.start_resolution = 64;
 }
 
 CCL_NAMESPACE_END

intern/cycles/app/cycles_xml.cpp

@@ -703,7 +703,7 @@ void xml_read_file(Scene *scene, const char *filepath)
 
   xml_read_include(state, path_filename(filepath));
 
-  scene->params.bvh_type = SceneParams::BVH_STATIC;
+  scene->params.bvh_type = BVH_TYPE_STATIC;
 }
 
 CCL_NAMESPACE_END

intern/cycles/blender/CMakeLists.txt

@@ -33,6 +33,7 @@ set(SRC
   blender_device.cpp
   blender_image.cpp
   blender_geometry.cpp
+  blender_gpu_display.cpp
   blender_light.cpp
   blender_mesh.cpp
   blender_object.cpp
@@ -50,6 +51,7 @@ set(SRC
 
   CCL_api.h
   blender_device.h
+  blender_gpu_display.h
   blender_id_map.h
   blender_image.h
   blender_object_cull.h
@@ -93,14 +95,9 @@ set(ADDON_FILES
 
 add_definitions(${GL_DEFINITIONS})
 
-if(WITH_CYCLES_DEVICE_OPENCL)
-  add_definitions(-DWITH_OPENCL)
+if(WITH_CYCLES_DEVICE_HIP)
+  add_definitions(-DWITH_HIP)
 endif()
-
-if(WITH_CYCLES_NETWORK)
-  add_definitions(-DWITH_NETWORK)
-endif()
-
 if(WITH_MOD_FLUID)
   add_definitions(-DWITH_FLUID)
 endif()

intern/cycles/blender/addon/__init__.py

@@ -58,7 +58,6 @@ class CyclesRender(bpy.types.RenderEngine):
     bl_use_eevee_viewport = True
     bl_use_preview = True
     bl_use_exclude_layers = True
-    bl_use_save_buffers = True
     bl_use_spherical_stereo = True
     bl_use_custom_freestyle = True
     bl_use_alembic_procedural = True
@@ -85,6 +84,12 @@ class CyclesRender(bpy.types.RenderEngine):
     def render(self, depsgraph):
         engine.render(self, depsgraph)
 
+    def render_frame_finish(self):
+        engine.render_frame_finish(self)
+
+    def draw(self, context, depsgraph):
+        engine.draw(self, depsgraph, context.space_data)
+
     def bake(self, depsgraph, obj, pass_type, pass_filter, width, height):
         engine.bake(self, depsgraph, obj, pass_type, pass_filter, width, height)
 
@@ -98,7 +103,7 @@ class CyclesRender(bpy.types.RenderEngine):
         engine.sync(self, depsgraph, context.blend_data)
 
     def view_draw(self, context, depsgraph):
-        engine.draw(self, depsgraph, context.region, context.space_data, context.region_data)
+        engine.view_draw(self, depsgraph, context.region, context.space_data, context.region_data)
 
     def update_script_node(self, node):
         if engine.with_osl():

intern/cycles/blender/addon/engine.py

@@ -18,62 +18,17 @@
 from __future__ import annotations
 
 
-def _is_using_buggy_driver():
-    import gpu
-    # 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.":
-        import re
-        version = gpu.platform.version_get()
-        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
-            # causing crashes.
-            return True
-        regex = re.compile(".*Compatibility Profile Context ([0-9]+(\\.[0-9]+)+)$")
-        if not regex.match(version):
-            # Skip cards like FireGL
-            return False
-        version = regex.sub("\\1", version).split('.')
-        return int(version[0]) == 8
-    return False
-
-
-def _workaround_buggy_drivers():
-    if _is_using_buggy_driver():
-        import _cycles
-        if hasattr(_cycles, "opencl_disable"):
-            print("Cycles: OpenGL driver known to be buggy, disabling OpenCL platform.")
-            _cycles.opencl_disable()
-
-
 def _configure_argument_parser():
     import argparse
     # No help because it conflicts with general Python scripts argument parsing
     parser = argparse.ArgumentParser(description="Cycles Addon argument parser",
                                      add_help=False)
-    parser.add_argument("--cycles-resumable-num-chunks",
-                        help="Number of chunks to split sample range into",
-                        default=None)
-    parser.add_argument("--cycles-resumable-current-chunk",
-                        help="Current chunk of samples range to render",
-                        default=None)
-    parser.add_argument("--cycles-resumable-start-chunk",
-                        help="Start chunk to render",
-                        default=None)
-    parser.add_argument("--cycles-resumable-end-chunk",
-                        help="End chunk to render",
-                        default=None)
     parser.add_argument("--cycles-print-stats",
                         help="Print rendering statistics to stderr",
                         action='store_true')
     parser.add_argument("--cycles-device",
                         help="Set the device to use for Cycles, overriding user preferences and the scene setting."
-                             "Valid options are 'CPU', 'CUDA', 'OPTIX' or 'OPENCL'."
+                             "Valid options are 'CPU', 'CUDA', 'OPTIX', or 'HIP'"
                              "Additionally, you can append '+CPU' to any GPU type for hybrid rendering.",
                         default=None)
     return parser
@@ -89,21 +44,6 @@ def _parse_command_line():
     parser = _configure_argument_parser()
     args, _ = parser.parse_known_args(argv[argv.index("--") + 1:])
 
-    if args.cycles_resumable_num_chunks is not None:
-        if args.cycles_resumable_current_chunk is not None:
-            import _cycles
-            _cycles.set_resumable_chunk(
-                int(args.cycles_resumable_num_chunks),
-                int(args.cycles_resumable_current_chunk),
-            )
-        elif args.cycles_resumable_start_chunk is not None and \
-                args.cycles_resumable_end_chunk:
-            import _cycles
-            _cycles.set_resumable_chunk_range(
-                int(args.cycles_resumable_num_chunks),
-                int(args.cycles_resumable_start_chunk),
-                int(args.cycles_resumable_end_chunk),
-            )
     if args.cycles_print_stats:
         import _cycles
         _cycles.enable_print_stats()
@@ -118,23 +58,11 @@ def init():
     import _cycles
     import os.path
 
-    # Workaround possibly buggy legacy drivers which crashes on the OpenCL
-    # device enumeration.
-    #
-    # This checks are not really correct because they might still fail
-    # in the case of multiple GPUs. However, currently buggy drivers
-    # are really old and likely to be used in single GPU systems only
-    # anyway.
-    #
-    # Can't do it in the background mode, so we hope OpenCL is no enabled
-    # in the user preferences.
-    if not bpy.app.background:
-        _workaround_buggy_drivers()
-
     path = os.path.dirname(__file__)
     user_path = os.path.dirname(os.path.abspath(bpy.utils.user_resource('CONFIG', path='')))
+    temp_path = bpy.app.tempdir
 
-    _cycles.init(path, user_path, bpy.app.background)
+    _cycles.init(path, user_path, temp_path, bpy.app.background)
     _parse_command_line()
 
 
@@ -177,6 +105,25 @@ def render(engine, depsgraph):
         _cycles.render(engine.session, depsgraph.as_pointer())
 
 
+def render_frame_finish(engine):
+    if not engine.session:
+        return
+
+    import _cycles
+    _cycles.render_frame_finish(engine.session)
+
+def draw(engine, depsgraph, space_image):
+    if not engine.session:
+        return
+
+    depsgraph_ptr = depsgraph.as_pointer()
+    space_image_ptr = space_image.as_pointer()
+    screen_ptr = space_image.id_data.as_pointer()
+
+    import _cycles
+    _cycles.draw(engine.session, depsgraph_ptr, screen_ptr, space_image_ptr)
+
+
 def bake(engine, depsgraph, obj, pass_type, pass_filter, width, height):
     import _cycles
     session = getattr(engine, "session", None)
@@ -204,14 +151,14 @@ def sync(engine, depsgraph, data):
     _cycles.sync(engine.session, depsgraph.as_pointer())
 
 
-def draw(engine, depsgraph, region, v3d, rv3d):
+def view_draw(engine, depsgraph, region, v3d, rv3d):
     import _cycles
     depsgraph = depsgraph.as_pointer()
     v3d = v3d.as_pointer()
     rv3d = rv3d.as_pointer()
 
     # draw render image
-    _cycles.draw(engine.session, depsgraph, v3d, rv3d)
+    _cycles.view_draw(engine.session, depsgraph, v3d, rv3d)
 
 
 def available_devices():
@@ -224,11 +171,6 @@ def with_osl():
     return _cycles.with_osl
 
 
-def with_network():
-    import _cycles
-    return _cycles.with_network
-
-
 def system_info():
     import _cycles
     return _cycles.system_info()
@@ -243,6 +185,7 @@ def list_render_passes(scene, srl):
     # Data passes.
     if srl.use_pass_z:                     yield ("Depth",         "Z",    'VALUE')
     if srl.use_pass_mist:                  yield ("Mist",          "Z",    'VALUE')
+    if srl.use_pass_position:              yield ("Position",      "XYZ",  'VECTOR')
     if srl.use_pass_normal:                yield ("Normal",        "XYZ",  'VECTOR')
     if srl.use_pass_vector:                yield ("Vector",        "XYZW", 'VECTOR')
     if srl.use_pass_uv:                    yield ("UV",            "UVA",  'VECTOR')
@@ -265,6 +208,7 @@ def list_render_passes(scene, srl):
     if srl.use_pass_environment:           yield ("Env",           "RGB",  'COLOR')
     if srl.use_pass_shadow:                yield ("Shadow",        "RGB",  'COLOR')
     if srl.use_pass_ambient_occlusion:     yield ("AO",            "RGB",  'COLOR')
+    if crl.use_pass_shadow_catcher:        yield ("Shadow Catcher",      "RGB",  'COLOR')
 
     # Debug passes.
     if crl.pass_debug_render_time:             yield ("Debug Render Time",             "X",   'VALUE')
@@ -283,30 +227,20 @@ def list_render_passes(scene, srl):
             yield ("CryptoAsset" + '{:02d}'.format(i), "RGBA", 'COLOR')
 
     # Denoising passes.
-    if (scene.cycles.use_denoising and crl.use_denoising) or crl.denoising_store_passes:
+    if scene.cycles.use_denoising and crl.use_denoising:
         yield ("Noisy Image", "RGBA", 'COLOR')
-        if crl.denoising_store_passes:
-            yield ("Denoising Normal",          "XYZ", 'VECTOR')
-            yield ("Denoising Albedo",          "RGB", 'COLOR')
-            yield ("Denoising Depth",           "Z",   'VALUE')
-
-            if scene.cycles.denoiser == 'NLM':
-                yield ("Denoising Shadowing",       "X",   'VALUE')
-                yield ("Denoising Variance",        "RGB", 'COLOR')
-                yield ("Denoising Intensity",       "X",   'VALUE')
-
-                clean_options = ("denoising_diffuse_direct", "denoising_diffuse_indirect",
-                                 "denoising_glossy_direct", "denoising_glossy_indirect",
-                                 "denoising_transmission_direct", "denoising_transmission_indirect")
-                if any(getattr(crl, option) for option in clean_options):
-                    yield ("Denoising Clean", "RGB", 'COLOR')
+        if crl.use_pass_shadow_catcher:
+            yield ("Noisy Shadow Catcher", "RGBA", 'COLOR')
+    if crl.denoising_store_passes:
+        yield ("Denoising Normal",          "XYZ", 'VECTOR')
+        yield ("Denoising Albedo",          "RGB", 'COLOR')
 
     # Custom AOV passes.
     for aov in srl.aovs:
         if aov.type == 'VALUE':
             yield (aov.name, "X", 'VALUE')
         else:
-            yield (aov.name, "RGBA", 'COLOR')
+            yield (aov.name, "RGB", 'COLOR')
 
 
 def register_passes(engine, scene, view_layer):

intern/cycles/blender/addon/presets.py

@@ -60,32 +60,48 @@ class AddPresetSampling(AddPresetBase, Operator):
     ]
 
     preset_values = [
+        "cycles.use_adaptive_sampling",
         "cycles.samples",
-        "cycles.preview_samples",
-        "cycles.aa_samples",
-        "cycles.preview_aa_samples",
-        "cycles.diffuse_samples",
-        "cycles.glossy_samples",
-        "cycles.transmission_samples",
-        "cycles.ao_samples",
-        "cycles.mesh_light_samples",
-        "cycles.subsurface_samples",
-        "cycles.volume_samples",
-        "cycles.use_square_samples",
-        "cycles.progressive",
-        "cycles.seed",
-        "cycles.sample_clamp_direct",
-        "cycles.sample_clamp_indirect",
-        "cycles.sample_all_lights_direct",
-        "cycles.sample_all_lights_indirect",
+        "cycles.adaptive_threshold",
+        "cycles.adaptive_min_samples",
+        "cycles.time_limit",
+        "cycles.use_denoising",
+        "cycles.denoiser",
+        "cycles.denoising_input_passes",
+        "cycles.denoising_prefilter",
     ]
 
     preset_subdir = "cycles/sampling"
 
 
+class AddPresetViewportSampling(AddPresetBase, Operator):
+    '''Add a Viewport Sampling Preset'''
+    bl_idname = "render.cycles_viewport_sampling_preset_add"
+    bl_label = "Add Viewport Sampling Preset"
+    preset_menu = "CYCLES_PT_viewport_sampling_presets"
+
+    preset_defines = [
+        "cycles = bpy.context.scene.cycles"
+    ]
+
+    preset_values = [
+        "cycles.use_preview_adaptive_sampling",
+        "cycles.preview_samples",
+        "cycles.preview_adaptive_threshold",
+        "cycles.preview_adaptive_min_samples",
+        "cycles.use_preview_denoising",
+        "cycles.preview_denoiser",
+        "cycles.preview_denoising_input_passes",
+        "cycles.preview_denoising_prefilter",
+        "cycles.preview_denoising_start_sample",
+    ]
+
+    preset_subdir = "cycles/viewport_sampling"
+
 classes = (
     AddPresetIntegrator,
     AddPresetSampling,
+    AddPresetViewportSampling,
 )
 
 

intern/cycles/blender/addon/properties.py

@@ -39,11 +39,6 @@ enum_devices = (
     ('GPU', "GPU Compute", "Use GPU compute device for rendering, configured in the system tab in the user preferences"),
 )
 
-from _cycles import with_network
-if with_network:
-    enum_devices += (('NETWORK', "Networked Device", "Use networked device for rendering"),)
-del with_network
-
 enum_feature_set = (
     ('SUPPORTED', "Supported", "Only use finished and supported features"),
     ('EXPERIMENTAL', "Experimental", "Use experimental and incomplete features that might be broken or change in the future", 'ERROR', 1),
@@ -84,15 +79,6 @@ enum_curve_shape = (
     ('THICK', "3D Curves", "Render hair as 3D curve, for accurate results when viewing hair close up"),
 )
 
-enum_tile_order = (
-    ('CENTER', "Center", "Render from center to the edges"),
-    ('RIGHT_TO_LEFT', "Right to Left", "Render from right to left"),
-    ('LEFT_TO_RIGHT', "Left to Right", "Render from left to right"),
-    ('TOP_TO_BOTTOM', "Top to Bottom", "Render from top to bottom"),
-    ('BOTTOM_TO_TOP', "Bottom to Top", "Render from bottom to top"),
-    ('HILBERT_SPIRAL', "Hilbert Spiral", "Render in a Hilbert Spiral"),
-)
-
 enum_use_layer_samples = (
     ('USE', "Use", "Per render layer number of samples override scene samples"),
     ('BOUNDED', "Bounded", "Bound per render layer number of samples by global samples"),
@@ -101,15 +87,9 @@ enum_use_layer_samples = (
 
 enum_sampling_pattern = (
     ('SOBOL', "Sobol", "Use Sobol random sampling pattern"),
-    ('CORRELATED_MUTI_JITTER', "Correlated Multi-Jitter", "Use Correlated Multi-Jitter random sampling pattern"),
     ('PROGRESSIVE_MUTI_JITTER', "Progressive Multi-Jitter", "Use Progressive Multi-Jitter random sampling pattern"),
 )
 
-enum_integrator = (
-    ('BRANCHED_PATH', "Branched Path Tracing", "Path tracing integrator that branches on the first bounce, giving more control over the number of light and material samples"),
-    ('PATH', "Path Tracing", "Pure path tracing integrator"),
-)
-
 enum_volume_sampling = (
     ('DISTANCE', "Distance", "Use distance sampling, best for dense volumes with lights far away"),
     ('EQUIANGULAR', "Equiangular", "Use equiangular sampling, best for volumes with low density with light inside or near the volume"),
@@ -131,7 +111,7 @@ enum_device_type = (
     ('CPU', "CPU", "CPU", 0),
     ('CUDA', "CUDA", "CUDA", 1),
     ('OPTIX', "OptiX", "OptiX", 3),
-    ('OPENCL', "OpenCL", "OpenCL", 2)
+    ("HIP", "HIP", "HIP", 4)
 )
 
 enum_texture_limit = (
@@ -145,38 +125,45 @@ enum_texture_limit = (
     ('8192', "8192", "Limit texture size to 8192 pixels", 7),
 )
 
+# NOTE: Identifiers are expected to be an upper case version of identifiers from  `Pass::get_type_enum()`
 enum_view3d_shading_render_pass = (
     ('', "General", ""),
 
-    ('COMBINED', "Combined", "Show the Combined Render pass", 1),
-    ('EMISSION', "Emission", "Show the Emission render pass", 33),
-    ('BACKGROUND', "Background", "Show the Background render pass", 34),
-    ('AO', "Ambient Occlusion", "Show the Ambient Occlusion render pass", 35),
+    ('COMBINED', "Combined", "Show the Combined Render pass"),
+    ('EMISSION', "Emission", "Show the Emission render pass"),
+    ('BACKGROUND', "Background", "Show the Background render pass"),
+    ('AO', "Ambient Occlusion", "Show the Ambient Occlusion render pass"),
+    ('SHADOW', "Shadow", "Show the Shadow render pass"),
+    ('SHADOW_CATCHER', "Shadow Catcher", "Show the Shadow Catcher render pass"),
 
     ('', "Light", ""),
 
-    ('DIFFUSE_DIRECT', "Diffuse Direct", "Show the Diffuse Direct render pass", 38),
-    ('DIFFUSE_INDIRECT', "Diffuse Indirect", "Show the Diffuse Indirect render pass", 39),
-    ('DIFFUSE_COLOR', "Diffuse Color", "Show the Diffuse Color render pass", 40),
+    ('DIFFUSE_DIRECT', "Diffuse Direct", "Show the Diffuse Direct render pass"),
+    ('DIFFUSE_INDIRECT', "Diffuse Indirect", "Show the Diffuse Indirect render pass"),
+    ('DIFFUSE_COLOR', "Diffuse Color", "Show the Diffuse Color render pass"),
 
-    ('GLOSSY_DIRECT', "Glossy Direct", "Show the Glossy Direct render pass", 41),
-    ('GLOSSY_INDIRECT', "Glossy Indirect", "Show the Glossy Indirect render pass", 42),
-    ('GLOSSY_COLOR', "Glossy Color", "Show the Glossy Color render pass", 43),
+    ('GLOSSY_DIRECT', "Glossy Direct", "Show the Glossy Direct render pass"),
+    ('GLOSSY_INDIRECT', "Glossy Indirect", "Show the Glossy Indirect render pass"),
+    ('GLOSSY_COLOR', "Glossy Color", "Show the Glossy Color render pass"),
 
     ('', "", ""),
 
-    ('TRANSMISSION_DIRECT', "Transmission Direct", "Show the Transmission Direct render pass", 44),
-    ('TRANSMISSION_INDIRECT', "Transmission Indirect", "Show the Transmission Indirect render pass", 45),
-    ('TRANSMISSION_COLOR', "Transmission Color", "Show the Transmission Color render pass", 46),
+    ('TRANSMISSION_DIRECT', "Transmission Direct", "Show the Transmission Direct render pass"),
+    ('TRANSMISSION_INDIRECT', "Transmission Indirect", "Show the Transmission Indirect render pass"),
+    ('TRANSMISSION_COLOR', "Transmission Color", "Show the Transmission Color render pass"),
 
-    ('VOLUME_DIRECT', "Volume Direct", "Show the Volume Direct render pass", 50),
-    ('VOLUME_INDIRECT', "Volume Indirect", "Show the Volume Indirect render pass", 51),
+    ('VOLUME_DIRECT', "Volume Direct", "Show the Volume Direct render pass"),
+    ('VOLUME_INDIRECT', "Volume Indirect", "Show the Volume Indirect render pass"),
 
     ('', "Data", ""),
 
-    ('NORMAL', "Normal", "Show the Normal render pass", 3),
-    ('UV', "UV", "Show the UV render pass", 4),
-    ('MIST', "Mist", "Show the Mist render pass", 32),
+    ('POSITION', "Position", "Show the Position render pass"),
+    ('NORMAL', "Normal", "Show the Normal render pass"),
+    ('UV', "UV", "Show the UV render pass"),
+    ('MIST', "Mist", "Show the Mist render pass"),
+    ('DENOISING_ALBEDO', "Denoising Albedo", "Albedo pass used by denoiser"),
+    ('DENOISING_NORMAL', "Denoising Normal", "Normal pass used by denoiser"),
+    ('SAMPLE_COUNT', "Sample Count", "Per-pixel number of samples"),
 )
 
 
@@ -208,18 +195,23 @@ def enum_preview_denoiser(self, context):
 
 
 def enum_denoiser(self, context):
-    items = [('NLM', "NLM", "Cycles native non-local means denoiser, running on any compute device", 1)]
+    items = []
     items += enum_optix_denoiser(self, context)
     items += enum_openimagedenoise_denoiser(self, context)
     return items
 
 
 enum_denoising_input_passes = (
-    ('RGB', "Color", "Use only color as input", 1),
-    ('RGB_ALBEDO', "Color + Albedo", "Use color and albedo data as input", 2),
-    ('RGB_ALBEDO_NORMAL', "Color + Albedo + Normal", "Use color, albedo and normal data as input", 3),
+    ('RGB', "None", "Don't use utility passes for denoising", 1),
+    ('RGB_ALBEDO', "Albedo", "Use albedo pass for denoising", 2),
+    ('RGB_ALBEDO_NORMAL', "Albedo and Normal", "Use albedo and normal passes for denoising", 3),
 )
 
+enum_denoising_prefilter = (
+    ('NONE', "None", "No prefiltering, use when guiding passes are noise-free", 1),
+    ('FAST', "Fast", "Denoise color and guiding passes together. Improves quality when guiding passes are noisy using least amount of extra processing time", 2),
+    ('ACCURATE', "Accurate", "Prefilter noisy guiding passes before denoising color. Improves quality when guiding passes are noisy using extra processing time", 3),
+)
 
 def update_render_passes(self, context):
     scene = context.scene
@@ -252,13 +244,6 @@ class CyclesRenderSettings(bpy.types.PropertyGroup):
         description="Use Open Shading Language (CPU rendering only)",
     )
 
-    progressive: EnumProperty(
-        name="Integrator",
-        description="Method to sample lights and materials",
-        items=enum_integrator,
-        default='PATH',
-    )
-
     preview_pause: BoolProperty(
         name="Pause Preview",
         description="Pause all viewport preview renders",
@@ -268,110 +253,88 @@ class CyclesRenderSettings(bpy.types.PropertyGroup):
     use_denoising: BoolProperty(
         name="Use Denoising",
         description="Denoise the rendered image",
-        default=False,
+        default=True,
         update=update_render_passes,
     )
+    denoiser: EnumProperty(
+        name="Denoiser",
+        description="Denoise the image with the selected denoiser. "
+        "For denoising the image after rendering",
+        items=enum_denoiser,
+        default=4, # Use integer to avoid error in builds without OpenImageDenoise.
+        update=update_render_passes,
+    )
+    denoising_prefilter: EnumProperty(
+        name="Denoising Prefilter",
+        description="Prefilter noisy guiding (albedo and normal) passes to improve denoising quality when using OpenImageDenoiser",
+        items=enum_denoising_prefilter,
+        default='ACCURATE',
+    )
+    denoising_input_passes: EnumProperty(
+        name="Denoising Input Passes",
+        description="Passes used by the denoiser to distinguish noise from shader and geometry detail",
+        items=enum_denoising_input_passes,
+        default='RGB_ALBEDO_NORMAL',
+    )
+
     use_preview_denoising: BoolProperty(
         name="Use Viewport Denoising",
         description="Denoise the image in the 3D viewport",
         default=False,
     )
-
-    denoiser: EnumProperty(
-        name="Denoiser",
-        description="Denoise the image with the selected denoiser. "
-        "For denoising the image after rendering, denoising data render passes "
-        "also adapt to the selected denoiser",
-        items=enum_denoiser,
-        default=1,
-        update=update_render_passes,
-    )
     preview_denoiser: EnumProperty(
         name="Viewport Denoiser",
         description="Denoise the image after each preview update with the selected denoiser",
         items=enum_preview_denoiser,
         default=0,
     )
-
-    use_square_samples: BoolProperty(
-        name="Square Samples",
-        description="Square sampling values for easier artist control",
-        default=False,
+    preview_denoising_prefilter: EnumProperty(
+        name="Viewport Denoising Prefilter",
+        description="Prefilter noisy guiding (albedo and normal) passes to improve denoising quality when using OpenImageDenoiser",
+        items=enum_denoising_prefilter,
+        default='FAST',
+    )
+    preview_denoising_input_passes: EnumProperty(
+        name="Viewport Denoising Input Passes",
+        description="Passes used by the denoiser to distinguish noise from shader and geometry detail",
+        items=enum_denoising_input_passes,
+        default='RGB_ALBEDO',
+    )
+    preview_denoising_start_sample: IntProperty(
+        name="Start Denoising",
+        description="Sample to start denoising the preview at",
+        min=0, max=(1 << 24),
+        default=1,
     )
 
     samples: IntProperty(
         name="Samples",
         description="Number of samples to render for each pixel",
         min=1, max=(1 << 24),
-        default=128,
+        default=4096,
     )
     preview_samples: IntProperty(
         name="Viewport Samples",
         description="Number of samples to render in the viewport, unlimited if 0",
         min=0, max=(1 << 24),
-        default=32,
-    )
-    aa_samples: IntProperty(
-        name="AA Samples",
-        description="Number of antialiasing samples to render for each pixel",
-        min=1, max=2097151,
-        default=128,
-    )
-    preview_aa_samples: IntProperty(
-        name="AA Samples",
-        description="Number of antialiasing samples to render in the viewport, unlimited if 0",
-        min=0, max=2097151,
-        default=32,
+        default=1024,
     )
 
-    diffuse_samples: IntProperty(
-        name="Diffuse Samples",
-        description="Number of diffuse bounce samples to render for each AA sample",
-        min=1, max=1024,
-        default=1,
-    )
-    glossy_samples: IntProperty(
-        name="Glossy Samples",
-        description="Number of glossy bounce samples to render for each AA sample",
-        min=1, max=1024,
-        default=1,
-    )
-    transmission_samples: IntProperty(
-        name="Transmission Samples",
-        description="Number of transmission bounce samples to render for each AA sample",
-        min=1, max=1024,
-        default=1,
-    )
-    ao_samples: IntProperty(
-        name="Ambient Occlusion Samples",
-        description="Number of ambient occlusion samples to render for each AA sample",
-        min=1, max=1024,
-        default=1,
-    )
-    mesh_light_samples: IntProperty(
-        name="Mesh Light Samples",
-        description="Number of mesh emission light samples to render for each AA sample",
-        min=1, max=1024,
-        default=1,
-    )
-    subsurface_samples: IntProperty(
-        name="Subsurface Samples",
-        description="Number of subsurface scattering samples to render for each AA sample",
-        min=1, max=1024,
-        default=1,
-    )
-    volume_samples: IntProperty(
-        name="Volume Samples",
-        description="Number of volume scattering samples to render for each AA sample",
-        min=1, max=1024,
-        default=1,
+    time_limit: FloatProperty(
+        name="Time Limit",
+        description="Limit the render time (excluding synchronization time)."
+        "Zero disables the limit",
+        min=0.0,
+        default=0.0,
+        step=100.0,
+        unit='TIME_ABSOLUTE',
     )
 
     sampling_pattern: EnumProperty(
         name="Sampling Pattern",
         description="Random sampling pattern used by the integrator",
         items=enum_sampling_pattern,
-        default='SOBOL',
+        default='PROGRESSIVE_MUTI_JITTER',
     )
 
     use_layer_samples: EnumProperty(
@@ -381,17 +344,6 @@ class CyclesRenderSettings(bpy.types.PropertyGroup):
         default='USE',
     )
 
-    sample_all_lights_direct: BoolProperty(
-        name="Sample All Direct Lights",
-        description="Sample all lights (for direct samples), rather than randomly picking one",
-        default=True,
-    )
-
-    sample_all_lights_indirect: BoolProperty(
-        name="Sample All Indirect Lights",
-        description="Sample all lights (for indirect samples), rather than randomly picking one",
-        default=True,
-    )
     light_sampling_threshold: FloatProperty(
         name="Light Sampling Threshold",
         description="Probabilistically terminate light samples when the light contribution is below this threshold (more noise but faster rendering). "
@@ -403,19 +355,39 @@ class CyclesRenderSettings(bpy.types.PropertyGroup):
     use_adaptive_sampling: BoolProperty(
         name="Use Adaptive Sampling",
         description="Automatically reduce the number of samples per pixel based on estimated noise level",
-        default=False,
+        default=True,
     )
-
     adaptive_threshold: FloatProperty(
         name="Adaptive Sampling Threshold",
         description="Noise level step to stop sampling at, lower values reduce noise at the cost of render time. Zero for automatic setting based on number of AA samples",
         min=0.0, max=1.0,
-        default=0.0,
+        soft_min=0.001,
+        default=0.01,
         precision=4,
     )
     adaptive_min_samples: IntProperty(
         name="Adaptive Min Samples",
-        description="Minimum AA samples for adaptive sampling, to discover noisy features before stopping sampling. Zero for automatic setting based on number of AA samples",
+        description="Minimum AA samples for adaptive sampling, to discover noisy features before stopping sampling. Zero for automatic setting based on noise threshold",
+        min=0, max=4096,
+        default=0,
+    )
+
+    use_preview_adaptive_sampling: BoolProperty(
+        name="Use Adaptive Sampling",
+        description="Automatically reduce the number of samples per pixel based on estimated noise level, for viewport renders",
+        default=True,
+    )
+    preview_adaptive_threshold: FloatProperty(
+        name="Adaptive Sampling Threshold",
+        description="Noise level step to stop sampling at, lower values reduce noise at the cost of render time. Zero for automatic setting based on number of AA samples, for viewport renders",
+        min=0.0, max=1.0,
+        soft_min=0.001,
+        default=0.1,
+        precision=4,
+    )
+    preview_adaptive_min_samples: IntProperty(
+        name="Adaptive Min Samples",
+        description="Minimum AA samples for adaptive sampling, to discover noisy features before stopping sampling. Zero for automatic setting based on noise threshold, for viewport renders",
         min=0, max=4096,
         default=0,
     )
@@ -632,53 +604,6 @@ class CyclesRenderSettings(bpy.types.PropertyGroup):
         default=10.0,
     )
 
-    debug_tile_size: IntProperty(
-        name="Tile Size",
-        description="",
-        min=1, max=4096,
-        default=1024,
-    )
-
-    preview_start_resolution: IntProperty(
-        name="Start Resolution",
-        description="Resolution to start rendering preview at, "
-        "progressively increasing it to the full viewport size",
-        min=8, max=16384,
-        default=64,
-        subtype='PIXEL'
-    )
-    preview_denoising_start_sample: IntProperty(
-        name="Start Denoising",
-        description="Sample to start denoising the preview at",
-        min=0, max=(1 << 24),
-        default=1,
-    )
-    preview_denoising_input_passes: EnumProperty(
-        name="Viewport Input Passes",
-        description="Passes used by the denoiser to distinguish noise from shader and geometry detail",
-        items=enum_denoising_input_passes,
-        default='RGB_ALBEDO',
-    )
-
-    debug_reset_timeout: FloatProperty(
-        name="Reset timeout",
-        description="",
-        min=0.01, max=10.0,
-        default=0.1,
-    )
-    debug_cancel_timeout: FloatProperty(
-        name="Cancel timeout",
-        description="",
-        min=0.01, max=10.0,
-        default=0.1,
-    )
-    debug_text_timeout: FloatProperty(
-        name="Text timeout",
-        description="",
-        min=0.01, max=10.0,
-        default=1.0,
-    )
-
     debug_bvh_type: EnumProperty(
         name="Viewport BVH Type",
         description="Choose between faster updates, or faster render",
@@ -701,38 +626,24 @@ class CyclesRenderSettings(bpy.types.PropertyGroup):
         default=0,
         min=0, max=16,
     )
-    tile_order: EnumProperty(
-        name="Tile Order",
-        description="Tile order for rendering",
-        items=enum_tile_order,
-        default='HILBERT_SPIRAL',
-        options=set(),  # Not animatable!
-    )
-    use_progressive_refine: BoolProperty(
-        name="Progressive Refine",
-        description="Instead of rendering each tile until it is finished, "
-        "refine the whole image progressively "
-        "(this renders somewhat slower, "
-        "but time can be saved by manually stopping the render when the noise is low enough)",
-        default=False,
-    )
 
     bake_type: EnumProperty(
         name="Bake Type",
         default='COMBINED',
         description="Type of pass to bake",
         items=(
-            ('COMBINED', "Combined", ""),
-            ('AO', "Ambient Occlusion", ""),
-            ('SHADOW', "Shadow", ""),
-            ('NORMAL', "Normal", ""),
-            ('UV', "UV", ""),
-            ('ROUGHNESS', "Roughness", ""),
-            ('EMIT', "Emit", ""),
-            ('ENVIRONMENT', "Environment", ""),
-            ('DIFFUSE', "Diffuse", ""),
-            ('GLOSSY', "Glossy", ""),
-            ('TRANSMISSION', "Transmission", ""),
+            ('COMBINED', "Combined", "", 0),
+            ('AO', "Ambient Occlusion", "", 1),
+            ('SHADOW', "Shadow", "", 2),
+            ('POSITION', "Position", "", 11),
+            ('NORMAL', "Normal", "", 3),
+            ('UV', "UV", "", 4),
+            ('ROUGHNESS', "Roughness", "", 5),
+            ('EMIT', "Emit", "", 6),
+            ('ENVIRONMENT', "Environment", "", 7),
+            ('DIFFUSE', "Diffuse", "", 8),
+            ('GLOSSY', "Glossy", "", 9),
+            ('TRANSMISSION', "Transmission", "", 10),
         ),
     )
 
@@ -827,6 +738,18 @@ class CyclesRenderSettings(bpy.types.PropertyGroup):
         min=0, max=1024,
     )
 
+    use_auto_tile: BoolProperty(
+        name="Auto Tiles",
+        description="Automatically render high resolution images in tiles to reduce memory usage, using the specified tile size. Tiles are cached to disk while rendering to save memory",
+        default=True,
+    )
+    tile_size: IntProperty(
+        name="Tile Size",
+        default=2048,
+        description="",
+        min=8, max=16384,
+    )
+
     # Various fine-tuning debug flags
 
     def _devices_update_callback(self, context):
@@ -844,45 +767,13 @@ class CyclesRenderSettings(bpy.types.PropertyGroup):
         items=enum_bvh_layouts,
         default='EMBREE',
     )
-    debug_use_cpu_split_kernel: BoolProperty(name="Split Kernel", default=False)
 
     debug_use_cuda_adaptive_compile: BoolProperty(name="Adaptive Compile", default=False)
-    debug_use_cuda_split_kernel: BoolProperty(name="Split Kernel", default=False)
 
-    debug_optix_cuda_streams: IntProperty(name="CUDA Streams", default=1, min=1)
-    debug_optix_curves_api: BoolProperty(name="Native OptiX Curve Primitive", default=False)
-
-    debug_opencl_kernel_type: EnumProperty(
-        name="OpenCL Kernel Type",
-        default='DEFAULT',
-        items=(
-            ('DEFAULT', "Default", ""),
-            ('MEGA', "Mega", ""),
-            ('SPLIT', "Split", ""),
-        ),
-        update=CyclesRenderSettings._devices_update_callback
-    )
-
-    debug_opencl_device_type: EnumProperty(
-        name="OpenCL Device Type",
-        default='ALL',
-        items=(
-            ('NONE', "None", ""),
-            ('ALL', "All", ""),
-            ('DEFAULT', "Default", ""),
-            ('CPU', "CPU", ""),
-            ('GPU', "GPU", ""),
-            ('ACCELERATOR', "Accelerator", ""),
-        ),
-        update=CyclesRenderSettings._devices_update_callback
-    )
-
-    debug_use_opencl_debug: BoolProperty(name="Debug OpenCL", default=False)
-
-    debug_opencl_mem_limit: IntProperty(
-        name="Memory limit",
-        default=0,
-        description="Artificial limit on OpenCL memory usage in MB (0 to disable limit)"
+    debug_use_optix_debug: BoolProperty(
+        name="OptiX Module Debug",
+        description="Load OptiX module in debug mode: lower logging verbosity level, enable validations, and lower optimization level",
+        default=False
     )
 
     @classmethod
@@ -1031,12 +922,6 @@ class CyclesLightSettings(bpy.types.PropertyGroup):
         description="Light casts shadows",
         default=True,
     )
-    samples: IntProperty(
-        name="Samples",
-        description="Number of light samples to render for each AA sample",
-        min=1, max=10000,
-        default=1,
-    )
     max_bounces: IntProperty(
         name="Max Bounces",
         description="Maximum number of bounces the light will contribute to the render",
@@ -1084,12 +969,6 @@ class CyclesWorldSettings(bpy.types.PropertyGroup):
         min=4, max=8192,
         default=1024,
     )
-    samples: IntProperty(
-        name="Samples",
-        description="Number of light samples to render for each AA sample",
-        min=1, max=10000,
-        default=1,
-    )
     max_bounces: IntProperty(
         name="Max Bounces",
         description="Maximum number of bounces the background light will contribute to the render",
@@ -1343,91 +1222,25 @@ class CyclesRenderLayerSettings(bpy.types.PropertyGroup):
         update=update_render_passes,
     )
 
+    use_pass_shadow_catcher: BoolProperty(
+        name="Shadow Catcher",
+        description="Pass containing shadows and light which is to be multiplied into backdrop",
+        default=False,
+        update=update_render_passes,
+    )
+
     use_denoising: BoolProperty(
         name="Use Denoising",
         description="Denoise the rendered image",
         default=True,
         update=update_render_passes,
     )
-    denoising_diffuse_direct: BoolProperty(
-        name="Diffuse Direct",
-        description="Denoise the direct diffuse lighting",
-        default=True,
-    )
-    denoising_diffuse_indirect: BoolProperty(
-        name="Diffuse Indirect",
-        description="Denoise the indirect diffuse lighting",
-        default=True,
-    )
-    denoising_glossy_direct: BoolProperty(
-        name="Glossy Direct",
-        description="Denoise the direct glossy lighting",
-        default=True,
-    )
-    denoising_glossy_indirect: BoolProperty(
-        name="Glossy Indirect",
-        description="Denoise the indirect glossy lighting",
-        default=True,
-    )
-    denoising_transmission_direct: BoolProperty(
-        name="Transmission Direct",
-        description="Denoise the direct transmission lighting",
-        default=True,
-    )
-    denoising_transmission_indirect: BoolProperty(
-        name="Transmission Indirect",
-        description="Denoise the indirect transmission lighting",
-        default=True,
-    )
-    denoising_strength: FloatProperty(
-        name="Denoising Strength",
-        description="Controls neighbor pixel weighting for the denoising filter (lower values preserve more detail, but aren't as smooth)",
-        min=0.0, max=1.0,
-        default=0.5,
-    )
-    denoising_feature_strength: FloatProperty(
-        name="Denoising Feature Strength",
-        description="Controls removal of noisy image feature passes (lower values preserve more detail, but aren't as smooth)",
-        min=0.0, max=1.0,
-        default=0.5,
-    )
-    denoising_radius: IntProperty(
-        name="Denoising Radius",
-        description="Size of the image area that's used to denoise a pixel (higher values are smoother, but might lose detail and are slower)",
-        min=1, max=25,
-        default=8,
-        subtype="PIXEL",
-    )
-    denoising_relative_pca: BoolProperty(
-        name="Relative Filter",
-        description="When removing pixels that don't carry information, use a relative threshold instead of an absolute one (can help to reduce artifacts, but might cause detail loss around edges)",
-        default=False,
-    )
     denoising_store_passes: BoolProperty(
         name="Store Denoising Passes",
         description="Store the denoising feature passes and the noisy image. The passes adapt to the denoiser selected for rendering",
         default=False,
         update=update_render_passes,
     )
-    denoising_neighbor_frames: IntProperty(
-        name="Neighbor Frames",
-        description="Number of neighboring frames to use for denoising animations (more frames produce smoother results at the cost of performance)",
-        min=0, max=7,
-        default=0,
-    )
-
-    denoising_optix_input_passes: EnumProperty(
-        name="Input Passes",
-        description="Passes used by the denoiser to distinguish noise from shader and geometry detail",
-        items=enum_denoising_input_passes,
-        default='RGB_ALBEDO',
-    )
-    denoising_openimagedenoise_input_passes: EnumProperty(
-        name="Input Passes",
-        description="Passes used by the denoiser to distinguish noise from shader and geometry detail",
-        items=enum_denoising_input_passes,
-        default='RGB_ALBEDO_NORMAL',
-    )
 
     @classmethod
     def register(cls):
@@ -1454,14 +1267,16 @@ class CyclesPreferences(bpy.types.AddonPreferences):
 
     def get_device_types(self, context):
         import _cycles
-        has_cuda, has_optix, has_opencl = _cycles.get_device_types()
+        has_cuda, has_optix, has_hip = _cycles.get_device_types()
+
         list = [('NONE', "None", "Don't use compute device", 0)]
         if has_cuda:
             list.append(('CUDA', "CUDA", "Use CUDA for GPU acceleration", 1))
         if has_optix:
             list.append(('OPTIX', "OptiX", "Use OptiX for GPU acceleration", 3))
-        if has_opencl:
-            list.append(('OPENCL', "OpenCL", "Use OpenCL for GPU acceleration", 2))
+        if has_hip:
+            list.append(('HIP', "HIP", "Use HIP for GPU acceleration", 4))
+
         return list
 
     compute_device_type: EnumProperty(
@@ -1486,7 +1301,7 @@ class CyclesPreferences(bpy.types.AddonPreferences):
 
     def update_device_entries(self, device_list):
         for device in device_list:
-            if not device[1] in {'CUDA', 'OPTIX', 'OPENCL', 'CPU'}:
+            if not device[1] in {'CUDA', 'OPTIX', 'CPU', 'HIP'}:
                 continue
             # Try to find existing Device entry
             entry = self.find_existing_device_entry(device)
@@ -1520,22 +1335,23 @@ class CyclesPreferences(bpy.types.AddonPreferences):
             elif entry.type == 'CPU':
                 cpu_devices.append(entry)
         # Extend all GPU devices with CPU.
-        if compute_device_type in {'CUDA', 'OPTIX', 'OPENCL'}:
+        if compute_device_type != 'CPU' and compute_device_type != 'HIP':
             devices.extend(cpu_devices)
         return devices
 
-    # For backwards compatibility, only returns CUDA and OpenCL but still
-    # refreshes all devices.
-    def get_devices(self, compute_device_type=''):
+    # Refresh device list. This does not happen automatically on Blender
+    # startup due to unstable OpenCL implementations that can cause crashes.
+    def refresh_devices(self):
         import _cycles
         # Ensure `self.devices` is not re-allocated when the second call to
         # get_devices_for_type is made, freeing items from the first list.
-        for device_type in ('CUDA', 'OPTIX', 'OPENCL'):
+        for device_type in ('CUDA', 'OPTIX', 'HIP'):
             self.update_device_entries(_cycles.available_devices(device_type))
 
-        cuda_devices = self.get_devices_for_type('CUDA')
-        opencl_devices = self.get_devices_for_type('OPENCL')
-        return cuda_devices, opencl_devices
+    # Deprecated: use refresh_devices instead.
+    def get_devices(self, compute_device_type=''):
+        self.refresh_devices()
+        return None
 
     def get_num_gpu_devices(self):
         import _cycles
@@ -1601,6 +1417,10 @@ class CyclesView3DShadingSettings(bpy.types.PropertyGroup):
         items=enum_view3d_shading_render_pass,
         default='COMBINED',
     )
+    show_active_pixels: BoolProperty(
+        name="Show Active Pixels",
+        description="When using adaptive sampling highlight pixels which are being sampled",
+    )
 
 
 def register():

intern/cycles/blender/addon/ui.py

@@ -34,6 +34,12 @@ class CYCLES_PT_sampling_presets(PresetPanel, Panel):
     preset_add_operator = "render.cycles_sampling_preset_add"
     COMPAT_ENGINES = {'CYCLES'}
 
+class CYCLES_PT_viewport_sampling_presets(PresetPanel, Panel):
+    bl_label = "Viewport Sampling Presets"
+    preset_subdir = "cycles/viewport_sampling"
+    preset_operator = "script.execute_preset"
+    preset_add_operator = "render.cycles_viewport_sampling_preset_add"
+    COMPAT_ENGINES = {'CYCLES'}
 
 class CYCLES_PT_integrator_presets(PresetPanel, Panel):
     bl_label = "Integrator Presets"
@@ -54,6 +60,15 @@ class CyclesButtonsPanel:
         return context.engine in cls.COMPAT_ENGINES
 
 
+class CyclesDebugButtonsPanel(CyclesButtonsPanel):
+    @classmethod
+    def poll(cls, context):
+        prefs = bpy.context.preferences
+        return (CyclesButtonsPanel.poll(context)
+                and prefs.experimental.use_cycles_debug
+                and prefs.view.show_developer_ui)
+
+
 # Adapt properties editor panel to display in node editor. We have to
 # copy the class rather than inherit due to the way bpy registration works.
 def node_panel(cls):
@@ -78,30 +93,23 @@ def use_cpu(context):
     return (get_device_type(context) == 'NONE' or cscene.device == 'CPU')
 
 
-def use_opencl(context):
-    cscene = context.scene.cycles
-
-    return (get_device_type(context) == 'OPENCL' and cscene.device == 'GPU')
-
-
 def use_cuda(context):
     cscene = context.scene.cycles
 
     return (get_device_type(context) == 'CUDA' and cscene.device == 'GPU')
 
 
+def use_hip(context):
+    cscene = context.scene.cycles
+
+    return (get_device_type(context) == 'HIP' and cscene.device == 'GPU')
+
 def use_optix(context):
     cscene = context.scene.cycles
 
     return (get_device_type(context) == 'OPTIX' and cscene.device == 'GPU')
 
 
-def use_branched_path(context):
-    cscene = context.scene.cycles
-
-    return (cscene.progressive == 'BRANCHED_PATH' and not use_optix(context))
-
-
 def use_sample_all_lights(context):
     cscene = context.scene.cycles
 
@@ -115,55 +123,93 @@ def show_device_active(context):
     return context.preferences.addons[__package__].preferences.has_active_device()
 
 
-def draw_samples_info(layout, context):
-    cscene = context.scene.cycles
-    integrator = cscene.progressive
+def get_effective_preview_denoiser(context):
+    scene = context.scene
+    cscene = scene.cycles
 
-    # Calculate sample values
-    if integrator == 'PATH':
-        aa = cscene.samples
-        if cscene.use_square_samples:
-            aa = aa * aa
-    else:
-        aa = cscene.aa_samples
-        d = cscene.diffuse_samples
-        g = cscene.glossy_samples
-        t = cscene.transmission_samples
-        ao = cscene.ao_samples
-        ml = cscene.mesh_light_samples
-        sss = cscene.subsurface_samples
-        vol = cscene.volume_samples
+    if cscene.preview_denoiser != "AUTO":
+        return cscene.preview_denoiser
 
-        if cscene.use_square_samples:
-            aa = aa * aa
-            d = d * d
-            g = g * g
-            t = t * t
-            ao = ao * ao
-            ml = ml * ml
-            sss = sss * sss
-            vol = vol * vol
+    if context.preferences.addons[__package__].preferences.get_devices_for_type('OPTIX'):
+        return 'OPTIX'
+
+    return 'OIDN'
 
-    # Draw interface
-    # Do not draw for progressive, when Square Samples are disabled
-    if use_branched_path(context) or (cscene.use_square_samples and integrator == 'PATH'):
-        col = layout.column(align=True)
-        col.scale_y = 0.6
-        col.label(text="Total Samples:")
-        col.separator()
-        if integrator == 'PATH':
-            col.label(text="%s AA" % aa)
-        else:
-            col.label(text="%s AA, %s Diffuse, %s Glossy, %s Transmission" %
-                      (aa, d * aa, g * aa, t * aa))
-            col.separator()
-            col.label(text="%s AO, %s Mesh Light, %s Subsurface, %s Volume" %
-                      (ao * aa, ml * aa, sss * aa, vol * aa))
 
 
 class CYCLES_RENDER_PT_sampling(CyclesButtonsPanel, Panel):
     bl_label = "Sampling"
 
+    def draw(self, context):
+        pass
+
+
+class CYCLES_RENDER_PT_sampling_viewport(CyclesButtonsPanel, Panel):
+    bl_label = "Viewport"
+    bl_parent_id = "CYCLES_RENDER_PT_sampling"
+
+    def draw_header_preset(self, context):
+        CYCLES_PT_viewport_sampling_presets.draw_panel_header(self.layout)
+
+    def draw(self, context):
+        layout = self.layout
+
+        scene = context.scene
+        cscene = scene.cycles
+
+        layout.use_property_split = True
+        layout.use_property_decorate = False
+
+        heading = layout.column(align=True, heading="Noise Threshold")
+        row = heading.row(align=True)
+        row.prop(cscene, "use_preview_adaptive_sampling", text="")
+        sub = row.row()
+        sub.active = cscene.use_preview_adaptive_sampling
+        sub.prop(cscene, "preview_adaptive_threshold", text="")
+
+        if cscene.use_preview_adaptive_sampling:
+            col = layout.column(align=True)
+            col.prop(cscene, "preview_samples", text=" Max Samples")
+            col.prop(cscene, "preview_adaptive_min_samples", text="Min Samples")
+        else:
+            layout.prop(cscene, "preview_samples", text="Samples")
+
+
+class CYCLES_RENDER_PT_sampling_viewport_denoise(CyclesButtonsPanel, Panel):
+    bl_label = "Denoise"
+    bl_parent_id = 'CYCLES_RENDER_PT_sampling_viewport'
+    bl_options = {'DEFAULT_CLOSED'}
+
+    def draw_header(self, context):
+        scene = context.scene
+        cscene = scene.cycles
+
+        self.layout.prop(context.scene.cycles, "use_preview_denoising", text="")
+
+    def draw(self, context):
+        layout = self.layout
+        layout.use_property_split = True
+        layout.use_property_decorate = False
+
+        scene = context.scene
+        cscene = scene.cycles
+
+        col = layout.column()
+        col.active = cscene.use_preview_denoising
+        col.prop(cscene, "preview_denoiser", text="Denoiser")
+        col.prop(cscene, "preview_denoising_input_passes", text="Passes")
+
+        effective_preview_denoiser = get_effective_preview_denoiser(context)
+        if effective_preview_denoiser == 'OPENIMAGEDENOISE':
+            col.prop(cscene, "preview_denoising_prefilter", text="Prefilter")
+
+        col.prop(cscene, "preview_denoising_start_sample", text="Start Sample")
+
+
+class CYCLES_RENDER_PT_sampling_render(CyclesButtonsPanel, Panel):
+    bl_label = "Render"
+    bl_parent_id = "CYCLES_RENDER_PT_sampling"
+
     def draw_header_preset(self, context):
         CYCLES_PT_sampling_presets.draw_panel_header(self.layout)
 
@@ -176,64 +222,32 @@ class CYCLES_RENDER_PT_sampling(CyclesButtonsPanel, Panel):
         layout.use_property_split = True
         layout.use_property_decorate = False
 
-        if not use_optix(context):
-            layout.prop(cscene, "progressive")
-
-        if not use_branched_path(context):
-            col = layout.column(align=True)
-            col.prop(cscene, "samples", text="Render")
-            col.prop(cscene, "preview_samples", text="Viewport")
-        else:
-            col = layout.column(align=True)
-            col.prop(cscene, "aa_samples", text="Render")
-            col.prop(cscene, "preview_aa_samples", text="Viewport")
-
-        if not use_branched_path(context):
-            draw_samples_info(layout, context)
-
-
-class CYCLES_RENDER_PT_sampling_sub_samples(CyclesButtonsPanel, Panel):
-    bl_label = "Sub Samples"
-    bl_parent_id = "CYCLES_RENDER_PT_sampling"
-
-    @classmethod
-    def poll(cls, context):
-        return use_branched_path(context)
-
-    def draw(self, context):
-        layout = self.layout
-        layout.use_property_split = True
-        layout.use_property_decorate = False
-
-        scene = context.scene
-        cscene = scene.cycles
+        heading = layout.column(align=True, heading="Noise Threshold")
+        row = heading.row(align=True)
+        row.prop(cscene, "use_adaptive_sampling", text="")
+        sub = row.row()
+        sub.active = cscene.use_adaptive_sampling
+        sub.prop(cscene, "adaptive_threshold", text="")
 
         col = layout.column(align=True)
-        col.prop(cscene, "diffuse_samples", text="Diffuse")
-        col.prop(cscene, "glossy_samples", text="Glossy")
-        col.prop(cscene, "transmission_samples", text="Transmission")
-        col.prop(cscene, "ao_samples", text="AO")
-
-        sub = col.row(align=True)
-        sub.active = use_sample_all_lights(context)
-        sub.prop(cscene, "mesh_light_samples", text="Mesh Light")
-        col.prop(cscene, "subsurface_samples", text="Subsurface")
-        col.prop(cscene, "volume_samples", text="Volume")
-
-        draw_samples_info(layout, context)
+        if cscene.use_adaptive_sampling:
+            col.prop(cscene, "samples", text=" Max Samples")
+            col.prop(cscene, "adaptive_min_samples", text="Min Samples")
+        else:
+            col.prop(cscene, "samples", text="Samples")
+        col.prop(cscene, "time_limit")
 
 
-class CYCLES_RENDER_PT_sampling_adaptive(CyclesButtonsPanel, Panel):
-    bl_label = "Adaptive Sampling"
-    bl_parent_id = "CYCLES_RENDER_PT_sampling"
+class CYCLES_RENDER_PT_sampling_render_denoise(CyclesButtonsPanel, Panel):
+    bl_label = "Denoise"
+    bl_parent_id = 'CYCLES_RENDER_PT_sampling_render'
     bl_options = {'DEFAULT_CLOSED'}
 
     def draw_header(self, context):
-        layout = self.layout
         scene = context.scene
         cscene = scene.cycles
 
-        layout.prop(cscene, "use_adaptive_sampling", text="")
+        self.layout.prop(context.scene.cycles, "use_denoising", text="")
 
     def draw(self, context):
         layout = self.layout
@@ -243,53 +257,12 @@ class CYCLES_RENDER_PT_sampling_adaptive(CyclesButtonsPanel, Panel):
         scene = context.scene
         cscene = scene.cycles
 
-        layout.active = cscene.use_adaptive_sampling
-
-        col = layout.column(align=True)
-        col.prop(cscene, "adaptive_threshold", text="Noise Threshold")
-        col.prop(cscene, "adaptive_min_samples", text="Min Samples")
-
-
-class CYCLES_RENDER_PT_sampling_denoising(CyclesButtonsPanel, Panel):
-    bl_label = "Denoising"
-    bl_parent_id = "CYCLES_RENDER_PT_sampling"
-    bl_options = {'DEFAULT_CLOSED'}
-
-    def draw(self, context):
-        layout = self.layout
-        layout.use_property_split = True
-        layout.use_property_decorate = False
-
-        scene = context.scene
-        cscene = scene.cycles
-
-        heading = layout.column(align=True, heading="Render")
-        row = heading.row(align=True)
-        row.prop(cscene, "use_denoising", text="")
-        sub = row.row()
-
-        sub.active = cscene.use_denoising
-        for view_layer in scene.view_layers:
-            if view_layer.cycles.denoising_store_passes:
-                sub.active = True
-
-        sub.prop(cscene, "denoiser", text="")
-
-        layout.separator()
-
-        heading = layout.column(align=False, heading="Viewport")
-        row = heading.row(align=True)
-        row.prop(cscene, "use_preview_denoising", text="")
-        sub = row.row()
-        sub.active = cscene.use_preview_denoising
-        sub.prop(cscene, "preview_denoiser", text="")
-
-        sub = heading.row(align=True)
-        sub.active = cscene.use_preview_denoising
-        sub.prop(cscene, "preview_denoising_start_sample", text="Start Sample")
-        sub = heading.row(align=True)
-        sub.active = cscene.use_preview_denoising
-        sub.prop(cscene, "preview_denoising_input_passes", text="Input Passes")
+        col = layout.column()
+        col.active = cscene.use_denoising
+        col.prop(cscene, "denoiser", text="Denoiser")
+        col.prop(cscene, "denoising_input_passes", text="Passes")
+        if cscene.denoiser == 'OPENIMAGEDENOISE':
+            col.prop(cscene, "denoising_prefilter", text="Prefilter")
 
 
 class CYCLES_RENDER_PT_sampling_advanced(CyclesButtonsPanel, Panel):
@@ -313,8 +286,6 @@ class CYCLES_RENDER_PT_sampling_advanced(CyclesButtonsPanel, Panel):
         col.active = not(cscene.use_adaptive_sampling)
         col.prop(cscene, "sampling_pattern", text="Pattern")
 
-        layout.prop(cscene, "use_square_samples")
-
         layout.separator()
 
         col = layout.column(align=True)
@@ -322,11 +293,6 @@ class CYCLES_RENDER_PT_sampling_advanced(CyclesButtonsPanel, Panel):
         col.prop(cscene, "min_transparent_bounces")
         col.prop(cscene, "light_sampling_threshold", text="Light Threshold")
 
-        if cscene.progressive != 'PATH' and use_branched_path(context):
-            col = layout.column(align=True)
-            col.prop(cscene, "sample_all_lights_direct")
-            col.prop(cscene, "sample_all_lights_indirect")
-
         for view_layer in scene.view_layers:
             if view_layer.samples > 0:
                 layout.separator()
@@ -334,62 +300,6 @@ class CYCLES_RENDER_PT_sampling_advanced(CyclesButtonsPanel, Panel):
                 break
 
 
-class CYCLES_RENDER_PT_sampling_total(CyclesButtonsPanel, Panel):
-    bl_label = "Total Samples"
-    bl_parent_id = "CYCLES_RENDER_PT_sampling"
-
-    @classmethod
-    def poll(cls, context):
-        scene = context.scene
-        cscene = scene.cycles
-
-        if cscene.use_square_samples:
-            return True
-
-        return cscene.progressive != 'PATH' and use_branched_path(context)
-
-    def draw(self, context):
-        layout = self.layout
-        cscene = context.scene.cycles
-        integrator = cscene.progressive
-
-        # Calculate sample values
-        if integrator == 'PATH':
-            aa = cscene.samples
-            if cscene.use_square_samples:
-                aa = aa * aa
-        else:
-            aa = cscene.aa_samples
-            d = cscene.diffuse_samples
-            g = cscene.glossy_samples
-            t = cscene.transmission_samples
-            ao = cscene.ao_samples
-            ml = cscene.mesh_light_samples
-            sss = cscene.subsurface_samples
-            vol = cscene.volume_samples
-
-            if cscene.use_square_samples:
-                aa = aa * aa
-                d = d * d
-                g = g * g
-                t = t * t
-                ao = ao * ao
-                ml = ml * ml
-                sss = sss * sss
-                vol = vol * vol
-
-        col = layout.column(align=True)
-        col.scale_y = 0.6
-        if integrator == 'PATH':
-            col.label(text="%s AA" % aa)
-        else:
-            col.label(text="%s AA, %s Diffuse, %s Glossy, %s Transmission" %
-                      (aa, d * aa, g * aa, t * aa))
-            col.separator()
-            col.label(text="%s AO, %s Mesh Light, %s Subsurface, %s Volume" %
-                      (ao * aa, ml * aa, sss * aa, vol * aa))
-
-
 class CYCLES_RENDER_PT_subdivision(CyclesButtonsPanel, Panel):
     bl_label = "Subdivision"
     bl_options = {'DEFAULT_CLOSED'}
@@ -548,6 +458,8 @@ class CYCLES_RENDER_PT_light_paths_fast_gi(CyclesButtonsPanel, Panel):
         layout.use_property_split = True
         layout.use_property_decorate = False
 
+        layout.active = cscene.use_fast_gi
+
         col = layout.column(align=True)
         col.prop(cscene, "ao_bounces", text="Viewport Bounces")
         col.prop(cscene, "ao_bounces_render", text="Render Bounces")
@@ -706,8 +618,8 @@ class CYCLES_RENDER_PT_performance_threads(CyclesButtonsPanel, Panel):
         sub.prop(rd, "threads")
 
 
-class CYCLES_RENDER_PT_performance_tiles(CyclesButtonsPanel, Panel):
-    bl_label = "Tiles"
+class CYCLES_RENDER_PT_performance_memory(CyclesButtonsPanel, Panel):
+    bl_label = "Memory"
     bl_parent_id = "CYCLES_RENDER_PT_performance"
 
     def draw(self, context):
@@ -716,19 +628,13 @@ class CYCLES_RENDER_PT_performance_tiles(CyclesButtonsPanel, Panel):
         layout.use_property_decorate = False
 
         scene = context.scene
-        rd = scene.render
         cscene = scene.cycles
 
         col = layout.column()
-
-        sub = col.column(align=True)
-        sub.prop(rd, "tile_x", text="Tiles X")
-        sub.prop(rd, "tile_y", text="Y")
-        col.prop(cscene, "tile_order", text="Order")
-
+        col.prop(cscene, "use_auto_tile")
         sub = col.column()
-        sub.active = not rd.use_save_buffers and not cscene.use_adaptive_sampling
-        sub.prop(cscene, "use_progressive_refine")
+        sub.active = cscene.use_auto_tile
+        sub.prop(cscene, "tile_size")
 
 
 class CYCLES_RENDER_PT_performance_acceleration_structure(CyclesButtonsPanel, Panel):
@@ -778,7 +684,6 @@ class CYCLES_RENDER_PT_performance_final_render(CyclesButtonsPanel, Panel):
 
         col = layout.column()
 
-        col.prop(rd, "use_save_buffers")
         col.prop(rd, "use_persistent_data", text="Persistent Data")
 
 
@@ -797,7 +702,6 @@ class CYCLES_RENDER_PT_performance_viewport(CyclesButtonsPanel, Panel):
 
         col = layout.column()
         col.prop(rd, "preview_pixel_size", text="Pixel Size")
-        col.prop(cscene, "preview_start_resolution", text="Start Pixels")
 
 
 class CYCLES_RENDER_PT_filter(CyclesButtonsPanel, Panel):
@@ -818,7 +722,6 @@ class CYCLES_RENDER_PT_filter(CyclesButtonsPanel, Panel):
 
         col = layout.column(heading="Include")
         col.prop(view_layer, "use_sky", text="Environment")
-        col.prop(view_layer, "use_ao", text="Ambient Occlusion")
         col.prop(view_layer, "use_solid", text="Surfaces")
         col.prop(view_layer, "use_strand", text="Hair")
         col.prop(view_layer, "use_volumes", text="Volumes")
@@ -827,6 +730,9 @@ class CYCLES_RENDER_PT_filter(CyclesButtonsPanel, Panel):
         sub = col.row()
         sub.prop(view_layer, "use_motion_blur", text="Motion Blur")
         sub.active = rd.use_motion_blur
+        sub = col.row()
+        sub.prop(view_layer.cycles, 'use_denoising', text='Denoising')
+        sub.active = scene.cycles.use_denoising
 
 
 class CYCLES_RENDER_PT_override(CyclesButtonsPanel, Panel):
@@ -872,6 +778,7 @@ class CYCLES_RENDER_PT_passes_data(CyclesButtonsPanel, Panel):
         col.prop(view_layer, "use_pass_combined")
         col.prop(view_layer, "use_pass_z")
         col.prop(view_layer, "use_pass_mist")
+        col.prop(view_layer, "use_pass_position")
         col.prop(view_layer, "use_pass_normal")
         sub = col.column()
         sub.active = not rd.use_motion_blur
@@ -928,6 +835,7 @@ class CYCLES_RENDER_PT_passes_light(CyclesButtonsPanel, Panel):
         col.prop(view_layer, "use_pass_environment")
         col.prop(view_layer, "use_pass_shadow")
         col.prop(view_layer, "use_pass_ambient_occlusion", text="Ambient Occlusion")
+        col.prop(cycles_view_layer, "use_pass_shadow_catcher")
 
 
 class CYCLES_RENDER_PT_passes_crypto(CyclesButtonsPanel, ViewLayerCryptomattePanel, Panel):
@@ -942,70 +850,6 @@ class CYCLES_RENDER_PT_passes_aov(CyclesButtonsPanel, ViewLayerAOVPanel):
     bl_parent_id = "CYCLES_RENDER_PT_passes"
 
 
-class CYCLES_RENDER_PT_denoising(CyclesButtonsPanel, Panel):
-    bl_label = "Denoising"
-    bl_context = "view_layer"
-    bl_options = {'DEFAULT_CLOSED'}
-
-    @classmethod
-    def poll(cls, context):
-        cscene = context.scene.cycles
-        return CyclesButtonsPanel.poll(context) and cscene.use_denoising
-
-    def draw_header(self, context):
-        scene = context.scene
-        view_layer = context.view_layer
-        cycles_view_layer = view_layer.cycles
-
-        layout = self.layout
-        layout.prop(cycles_view_layer, "use_denoising", text="")
-
-    def draw(self, context):
-        layout = self.layout
-        layout.use_property_split = True
-        layout.use_property_decorate = False
-
-        scene = context.scene
-        view_layer = context.view_layer
-        cycles_view_layer = view_layer.cycles
-        denoiser = scene.cycles.denoiser
-
-        layout.active = denoiser != 'NONE' and cycles_view_layer.use_denoising
-
-        col = layout.column()
-
-        if denoiser == 'OPTIX':
-            col.prop(cycles_view_layer, "denoising_optix_input_passes")
-            return
-        elif denoiser == 'OPENIMAGEDENOISE':
-            col.prop(cycles_view_layer, "denoising_openimagedenoise_input_passes")
-            return
-
-        col.prop(cycles_view_layer, "denoising_radius", text="Radius")
-
-        col = layout.column()
-        col.prop(cycles_view_layer, "denoising_strength", slider=True, text="Strength")
-        col.prop(cycles_view_layer, "denoising_feature_strength", slider=True, text="Feature Strength")
-        col.prop(cycles_view_layer, "denoising_relative_pca")
-
-        layout.separator()
-
-        col = layout.column()
-        col.active = cycles_view_layer.use_denoising or cycles_view_layer.denoising_store_passes
-
-        row = col.row(heading="Diffuse", align=True)
-        row.prop(cycles_view_layer, "denoising_diffuse_direct", text="Direct", toggle=True)
-        row.prop(cycles_view_layer, "denoising_diffuse_indirect", text="Indirect", toggle=True)
-
-        row = col.row(heading="Glossy", align=True)
-        row.prop(cycles_view_layer, "denoising_glossy_direct", text="Direct", toggle=True)
-        row.prop(cycles_view_layer, "denoising_glossy_indirect", text="Indirect", toggle=True)
-
-        row = col.row(heading="Transmission", align=True)
-        row.prop(cycles_view_layer, "denoising_transmission_direct", text="Direct", toggle=True)
-        row.prop(cycles_view_layer, "denoising_transmission_indirect", text="Indirect", toggle=True)
-
-
 class CYCLES_PT_post_processing(CyclesButtonsPanel, Panel):
     bl_label = "Post Processing"
     bl_options = {'DEFAULT_CLOSED'}
@@ -1417,10 +1261,6 @@ class CYCLES_LIGHT_PT_light(CyclesButtonsPanel, Panel):
 
         if not (light.type == 'AREA' and clamp.is_portal):
             sub = col.column()
-            if use_branched_path(context):
-                subsub = sub.row(align=True)
-                subsub.active = use_sample_all_lights(context)
-                subsub.prop(clamp, "samples")
             sub.prop(clamp, "max_bounces")
 
         sub = col.column(align=True)
@@ -1526,34 +1366,6 @@ class CYCLES_WORLD_PT_volume(CyclesButtonsPanel, Panel):
         panel_node_draw(layout, world, 'OUTPUT_WORLD', 'Volume')
 
 
-class CYCLES_WORLD_PT_ambient_occlusion(CyclesButtonsPanel, Panel):
-    bl_label = "Ambient Occlusion"
-    bl_context = "world"
-    bl_options = {'DEFAULT_CLOSED'}
-
-    @classmethod
-    def poll(cls, context):
-        return context.world and CyclesButtonsPanel.poll(context)
-
-    def draw_header(self, context):
-        light = context.world.light_settings
-        self.layout.prop(light, "use_ambient_occlusion", text="")
-
-    def draw(self, context):
-        layout = self.layout
-        layout.use_property_split = True
-        layout.use_property_decorate = False
-
-        light = context.world.light_settings
-        scene = context.scene
-
-        col = layout.column()
-        sub = col.column()
-        sub.active = light.use_ambient_occlusion or scene.render.use_simplify
-        sub.prop(light, "ao_factor", text="Factor")
-        col.prop(light, "distance", text="Distance")
-
-
 class CYCLES_WORLD_PT_mist(CyclesButtonsPanel, Panel):
     bl_label = "Mist Pass"
     bl_context = "world"
@@ -1650,10 +1462,6 @@ class CYCLES_WORLD_PT_settings_surface(CyclesButtonsPanel, Panel):
         subsub = sub.row(align=True)
         subsub.active = cworld.sampling_method == 'MANUAL'
         subsub.prop(cworld, "sample_map_resolution")
-        if use_branched_path(context):
-            subsub = sub.column(align=True)
-            subsub.active = use_sample_all_lights(context)
-            subsub.prop(cworld, "samples")
         sub.prop(cworld, "max_bounces")
 
 
@@ -1677,8 +1485,7 @@ class CYCLES_WORLD_PT_settings_volume(CyclesButtonsPanel, Panel):
         col = layout.column()
 
         sub = col.column()
-        sub.active = use_cpu(context)
-        sub.prop(cworld, "volume_sampling", text="Sampling")
+        col.prop(cworld, "volume_sampling", text="Sampling")
         col.prop(cworld, "volume_interpolation", text="Interpolation")
         col.prop(cworld, "homogeneous_volume", text="Homogeneous")
         sub = col.column()
@@ -1817,8 +1624,7 @@ class CYCLES_MATERIAL_PT_settings_volume(CyclesButtonsPanel, Panel):
 
         col = layout.column()
         sub = col.column()
-        sub.active = use_cpu(context)
-        sub.prop(cmat, "volume_sampling", text="Sampling")
+        col.prop(cmat, "volume_sampling", text="Sampling")
         col.prop(cmat, "volume_interpolation", text="Interpolation")
         col.prop(cmat, "homogeneous_volume", text="Homogeneous")
         sub = col.column()
@@ -1845,9 +1651,6 @@ class CYCLES_RENDER_PT_bake(CyclesButtonsPanel, Panel):
         cbk = scene.render.bake
         rd = scene.render
 
-        if use_optix(context):
-            layout.label(text="Baking is performed using CUDA instead of OptiX", icon='INFO')
-
         if rd.use_bake_multires:
             layout.operator("object.bake_image", icon='RENDER_STILL')
             layout.prop(rd, "use_bake_multires")
@@ -1905,7 +1708,6 @@ class CYCLES_RENDER_PT_bake_influence(CyclesButtonsPanel, Panel):
             col.prop(cbk, "use_pass_diffuse")
             col.prop(cbk, "use_pass_glossy")
             col.prop(cbk, "use_pass_transmission")
-            col.prop(cbk, "use_pass_ambient_occlusion")
             col.prop(cbk, "use_pass_emit")
 
         elif cscene.bake_type in {'DIFFUSE', 'GLOSSY', 'TRANSMISSION'}:
@@ -1989,19 +1791,12 @@ class CYCLES_RENDER_PT_bake_output(CyclesButtonsPanel, Panel):
                 layout.prop(cbk, "use_clear", text="Clear Image")
 
 
-class CYCLES_RENDER_PT_debug(CyclesButtonsPanel, Panel):
+class CYCLES_RENDER_PT_debug(CyclesDebugButtonsPanel, Panel):
     bl_label = "Debug"
     bl_context = "render"
     bl_options = {'DEFAULT_CLOSED'}
     COMPAT_ENGINES = {'CYCLES'}
 
-    @classmethod
-    def poll(cls, context):
-        prefs = bpy.context.preferences
-        return (CyclesButtonsPanel.poll(context)
-                and prefs.experimental.use_cycles_debug
-                and prefs.view.show_developer_ui)
-
     def draw(self, context):
         layout = self.layout
 
@@ -2018,29 +1813,18 @@ class CYCLES_RENDER_PT_debug(CyclesButtonsPanel, Panel):
         row.prop(cscene, "debug_use_cpu_avx", toggle=True)
         row.prop(cscene, "debug_use_cpu_avx2", toggle=True)
         col.prop(cscene, "debug_bvh_layout")
-        col.prop(cscene, "debug_use_cpu_split_kernel")
 
         col.separator()
 
         col = layout.column()
         col.label(text="CUDA Flags:")
         col.prop(cscene, "debug_use_cuda_adaptive_compile")
-        col.prop(cscene, "debug_use_cuda_split_kernel")
 
         col.separator()
 
         col = layout.column()
         col.label(text="OptiX Flags:")
-        col.prop(cscene, "debug_optix_cuda_streams")
-        col.prop(cscene, "debug_optix_curves_api")
-
-        col.separator()
-
-        col = layout.column()
-        col.label(text="OpenCL Flags:")
-        col.prop(cscene, "debug_opencl_device_type", text="Device")
-        col.prop(cscene, "debug_use_opencl_debug", text="Debug")
-        col.prop(cscene, "debug_opencl_mem_limit")
+        col.prop(cscene, "debug_use_optix_debug")
 
         col.separator()
 
@@ -2141,20 +1925,22 @@ class CYCLES_RENDER_PT_simplify_culling(CyclesButtonsPanel, Panel):
         sub.prop(cscene, "distance_cull_margin", text="")
 
 
-class CYCLES_VIEW3D_PT_shading_render_pass(Panel):
+class CyclesShadingButtonsPanel(CyclesButtonsPanel):
     bl_space_type = 'VIEW_3D'
     bl_region_type = 'HEADER'
-    bl_label = "Render Pass"
     bl_parent_id = 'VIEW3D_PT_shading'
-    COMPAT_ENGINES = {'CYCLES'}
 
     @classmethod
     def poll(cls, context):
         return (
-            context.engine in cls.COMPAT_ENGINES and
+            CyclesButtonsPanel.poll(context) and
             context.space_data.shading.type == 'RENDERED'
         )
 
+
+class CYCLES_VIEW3D_PT_shading_render_pass(CyclesShadingButtonsPanel, Panel):
+    bl_label = "Render Pass"
+
     def draw(self, context):
         shading = context.space_data.shading
 
@@ -2162,6 +1948,26 @@ class CYCLES_VIEW3D_PT_shading_render_pass(Panel):
         layout.prop(shading.cycles, "render_pass", text="")
 
 
+class CYCLES_VIEW3D_PT_shading_debug(CyclesDebugButtonsPanel,
+                                     CyclesShadingButtonsPanel,
+                                     Panel):
+    bl_label = "Debug"
+
+    @classmethod
+    def poll(cls, context):
+        return (
+            CyclesDebugButtonsPanel.poll(context) and
+            CyclesShadingButtonsPanel.poll(context)
+        )
+
+    def draw(self, context):
+        shading = context.space_data.shading
+
+        layout = self.layout
+        layout.active = context.scene.cycles.use_preview_adaptive_sampling
+        layout.prop(shading.cycles, "show_active_pixels")
+
+
 class CYCLES_VIEW3D_PT_shading_lighting(Panel):
     bl_space_type = 'VIEW_3D'
     bl_region_type = 'HEADER'
@@ -2275,11 +2081,13 @@ def get_panels():
 
 classes = (
     CYCLES_PT_sampling_presets,
+    CYCLES_PT_viewport_sampling_presets,
     CYCLES_PT_integrator_presets,
     CYCLES_RENDER_PT_sampling,
-    CYCLES_RENDER_PT_sampling_sub_samples,
-    CYCLES_RENDER_PT_sampling_adaptive,
-    CYCLES_RENDER_PT_sampling_denoising,
+    CYCLES_RENDER_PT_sampling_viewport,
+    CYCLES_RENDER_PT_sampling_viewport_denoise,
+    CYCLES_RENDER_PT_sampling_render,
+    CYCLES_RENDER_PT_sampling_render_denoise,
     CYCLES_RENDER_PT_sampling_advanced,
     CYCLES_RENDER_PT_light_paths,
     CYCLES_RENDER_PT_light_paths_max_bounces,
@@ -2296,6 +2104,7 @@ classes = (
     CYCLES_VIEW3D_PT_simplify_greasepencil,
     CYCLES_VIEW3D_PT_shading_lighting,
     CYCLES_VIEW3D_PT_shading_render_pass,
+    CYCLES_VIEW3D_PT_shading_debug,
     CYCLES_RENDER_PT_motion_blur,
     CYCLES_RENDER_PT_motion_blur_curve,
     CYCLES_RENDER_PT_film,
@@ -2303,7 +2112,7 @@ classes = (
     CYCLES_RENDER_PT_film_transparency,
     CYCLES_RENDER_PT_performance,
     CYCLES_RENDER_PT_performance_threads,
-    CYCLES_RENDER_PT_performance_tiles,
+    CYCLES_RENDER_PT_performance_memory,
     CYCLES_RENDER_PT_performance_acceleration_structure,
     CYCLES_RENDER_PT_performance_final_render,
     CYCLES_RENDER_PT_performance_viewport,
@@ -2314,7 +2123,6 @@ classes = (
     CYCLES_RENDER_PT_passes_aov,
     CYCLES_RENDER_PT_filter,
     CYCLES_RENDER_PT_override,
-    CYCLES_RENDER_PT_denoising,
     CYCLES_PT_post_processing,
     CYCLES_CAMERA_PT_dof,
     CYCLES_CAMERA_PT_dof_aperture,
@@ -2333,7 +2141,6 @@ classes = (
     CYCLES_WORLD_PT_preview,
     CYCLES_WORLD_PT_surface,
     CYCLES_WORLD_PT_volume,
-    CYCLES_WORLD_PT_ambient_occlusion,
     CYCLES_WORLD_PT_mist,
     CYCLES_WORLD_PT_ray_visibility,
     CYCLES_WORLD_PT_settings,

intern/cycles/blender/addon/version_update.py

@@ -109,7 +109,7 @@ def do_versions(self):
         library_versions.setdefault(library.version, []).append(library)
 
     # Do versioning per library, since they might have different versions.
-    max_need_versioning = (2, 93, 7)
+    max_need_versioning = (3, 0, 25)
     for version, libraries in library_versions.items():
         if version > max_need_versioning:
             continue
@@ -166,10 +166,6 @@ def do_versions(self):
                 if not cscene.is_property_set("filter_type"):
                     cscene.pixel_filter_type = 'GAUSSIAN'
 
-                # Tile Order
-                if not cscene.is_property_set("tile_order"):
-                    cscene.tile_order = 'CENTER'
-
             if version <= (2, 76, 10):
                 cscene = scene.cycles
                 if cscene.is_property_set("filter_type"):
@@ -186,10 +182,6 @@ def do_versions(self):
             if version <= (2, 79, 0):
                 cscene = scene.cycles
                 # Default changes
-                if not cscene.is_property_set("aa_samples"):
-                    cscene.aa_samples = 4
-                if not cscene.is_property_set("preview_aa_samples"):
-                    cscene.preview_aa_samples = 4
                 if not cscene.is_property_set("blur_glossy"):
                     cscene.blur_glossy = 0.0
                 if not cscene.is_property_set("sample_clamp_indirect"):
@@ -203,7 +195,6 @@ def do_versions(self):
                     view_layer.use_pass_cryptomatte_material = cview_layer.get("use_pass_crypto_material", False)
                     view_layer.use_pass_cryptomatte_asset = cview_layer.get("use_pass_crypto_asset", False)
                     view_layer.pass_cryptomatte_depth = cview_layer.get("pass_crypto_depth", 6)
-                    view_layer.use_pass_cryptomatte_accurate = cview_layer.get("pass_crypto_accurate", True)
 
             if version <= (2, 93, 7):
                 if scene.render.engine == 'CYCLES':
@@ -229,6 +220,35 @@ def do_versions(self):
                     cscene.ao_bounces = 1
                     cscene.ao_bounces_render = 1
 
+            if version <= (3, 0, 25):
+                cscene = scene.cycles
+
+                # Default changes.
+                if not cscene.is_property_set("samples"):
+                    cscene.samples = 128
+                if not cscene.is_property_set("preview_samples"):
+                    cscene.preview_samples = 32
+                if not cscene.is_property_set("use_adaptive_sampling"):
+                    cscene.use_adaptive_sampling = False
+                    cscene.use_preview_adaptive_sampling = False
+                if not cscene.is_property_set("use_denoising"):
+                    cscene.use_denoising = False
+                if not cscene.is_property_set("use_preview_denoising"):
+                    cscene.use_preview_denoising = False
+                if not cscene.is_property_set("sampling_pattern"):
+                    cscene.sampling_pattern = 'PROGRESSIVE_MUTI_JITTER'
+
+                # Removal of square samples.
+                cscene = scene.cycles
+                use_square_samples = cscene.get("use_square_samples", False)
+
+                if use_square_samples:
+                    cscene.samples *= cscene.samples
+                    cscene.preview_samples *= cscene.preview_samples
+                    for layer in scene.view_layers:
+                        layer.samples *= layer.samples
+                    cscene["use_square_samples"] = False
+
         # Lamps
         for light in bpy.data.lights:
             if light.library not in libraries:
@@ -249,10 +269,6 @@ def do_versions(self):
             if version <= (2, 76, 9):
                 cworld = world.cycles
 
-                # World MIS Samples
-                if not cworld.is_property_set("samples"):
-                    cworld.samples = 4
-
                 # World MIS Resolution
                 if not cworld.is_property_set("sample_map_resolution"):
                     cworld.sample_map_resolution = 256

intern/cycles/blender/blender_camera.cpp

@@ -894,12 +894,8 @@ void BlenderSync::sync_view(BL::SpaceView3D &b_v3d,
   }
 }
 
-BufferParams BlenderSync::get_buffer_params(BL::SpaceView3D &b_v3d,
-                                            BL::RegionView3D &b_rv3d,
-                                            Camera *cam,
-                                            int width,
-                                            int height,
-                                            const bool use_denoiser)
+BufferParams BlenderSync::get_buffer_params(
+    BL::SpaceView3D &b_v3d, BL::RegionView3D &b_rv3d, Camera *cam, int width, int height)
 {
   BufferParams params;
   bool use_border = false;
@@ -931,11 +927,6 @@ BufferParams BlenderSync::get_buffer_params(BL::SpaceView3D &b_v3d,
     params.height = height;
   }
 
-  PassType display_pass = update_viewport_display_passes(b_v3d, params.passes);
-
-  /* Can only denoise the combined image pass */
-  params.denoising_data_pass = display_pass == PASS_COMBINED && use_denoiser;
-
   return params;
 }
 

intern/cycles/blender/blender_curves.cpp

@@ -283,10 +283,13 @@ static void ExportCurveSegments(Scene *scene, Hair *hair, ParticleCurveData *CDa
     return;
 
   Attribute *attr_intercept = NULL;
+  Attribute *attr_length = NULL;
   Attribute *attr_random = NULL;
 
   if (hair->need_attribute(scene, ATTR_STD_CURVE_INTERCEPT))
     attr_intercept = hair->attributes.add(ATTR_STD_CURVE_INTERCEPT);
+  if (hair->need_attribute(scene, ATTR_STD_CURVE_LENGTH))
+    attr_length = hair->attributes.add(ATTR_STD_CURVE_LENGTH);
   if (hair->need_attribute(scene, ATTR_STD_CURVE_RANDOM))
     attr_random = hair->attributes.add(ATTR_STD_CURVE_RANDOM);
 
@@ -336,6 +339,10 @@ static void ExportCurveSegments(Scene *scene, Hair *hair, ParticleCurveData *CDa
         num_curve_keys++;
       }
 
+      if (attr_length != NULL) {
+        attr_length->add(CData->curve_length[curve]);
+      }
+
       if (attr_random != NULL) {
         attr_random->add(hash_uint2_to_float(num_curves, 0));
       }
@@ -657,11 +664,15 @@ static void export_hair_curves(Scene *scene, Hair *hair, BL::Hair b_hair)
 
   /* Add requested attributes. */
   Attribute *attr_intercept = NULL;
+  Attribute *attr_length = NULL;
   Attribute *attr_random = NULL;
 
   if (hair->need_attribute(scene, ATTR_STD_CURVE_INTERCEPT)) {
     attr_intercept = hair->attributes.add(ATTR_STD_CURVE_INTERCEPT);
   }
+  if (hair->need_attribute(scene, ATTR_STD_CURVE_LENGTH)) {
+    attr_length = hair->attributes.add(ATTR_STD_CURVE_LENGTH);
+  }
   if (hair->need_attribute(scene, ATTR_STD_CURVE_RANDOM)) {
     attr_random = hair->attributes.add(ATTR_STD_CURVE_RANDOM);
   }
@@ -714,6 +725,10 @@ static void export_hair_curves(Scene *scene, Hair *hair, BL::Hair b_hair)
       }
     }
 
+    if (attr_length) {
+      attr_length->add(length);
+    }
+
     /* Random number per curve. */
     if (attr_random != NULL) {
       attr_random->add(hash_uint2_to_float(b_curve.index(), 0));

intern/cycles/blender/blender_device.cpp

@@ -25,8 +25,8 @@ CCL_NAMESPACE_BEGIN
 enum ComputeDevice {
   COMPUTE_DEVICE_CPU = 0,
   COMPUTE_DEVICE_CUDA = 1,
-  COMPUTE_DEVICE_OPENCL = 2,
   COMPUTE_DEVICE_OPTIX = 3,
+  COMPUTE_DEVICE_HIP = 4,
 
   COMPUTE_DEVICE_NUM
 };
@@ -68,13 +68,6 @@ DeviceInfo blender_device_info(BL::Preferences &b_preferences, BL::Scene &b_scen
       device = Device::get_multi_device(devices, threads, background);
     }
   }
-  else if (get_enum(cscene, "device") == 2) {
-    /* Find network device. */
-    vector<DeviceInfo> devices = Device::available_devices(DEVICE_MASK_NETWORK);
-    if (!devices.empty()) {
-      device = devices.front();
-    }
-  }
   else if (get_enum(cscene, "device") == 1) {
     /* Test if we are using GPU devices. */
     ComputeDevice compute_device = (ComputeDevice)get_enum(
@@ -89,8 +82,8 @@ DeviceInfo blender_device_info(BL::Preferences &b_preferences, BL::Scene &b_scen
       else if (compute_device == COMPUTE_DEVICE_OPTIX) {
         mask |= DEVICE_MASK_OPTIX;
       }
-      else if (compute_device == COMPUTE_DEVICE_OPENCL) {
-        mask |= DEVICE_MASK_OPENCL;
+      else if (compute_device == COMPUTE_DEVICE_HIP) {
+        mask |= DEVICE_MASK_HIP;
       }
       vector<DeviceInfo> devices = Device::available_devices(mask);
 

intern/cycles/blender/blender_geometry.cpp

@@ -80,7 +80,11 @@ Geometry *BlenderSync::sync_geometry(BL::Depsgraph &b_depsgraph,
 {
   /* Test if we can instance or if the object is modified. */
   Geometry::Type geom_type = determine_geom_type(b_ob_info, use_particle_hair);
-  GeometryKey key(b_ob_info.object_data, geom_type);
+  BL::ID b_key_id = (b_ob_info.is_real_object_data() &&
+                     BKE_object_is_modified(b_ob_info.real_object)) ?
+                        b_ob_info.real_object :
+                        b_ob_info.object_data;
+  GeometryKey key(b_key_id.ptr.data, geom_type);
 
   /* Find shader indices. */
   array<Node *> used_shaders = find_used_shaders(b_ob_info.iter_object);
@@ -110,7 +114,7 @@ Geometry *BlenderSync::sync_geometry(BL::Depsgraph &b_depsgraph,
   }
   else {
     /* Test if we need to update existing geometry. */
-    sync = geometry_map.update(geom, b_ob_info.object_data);
+    sync = geometry_map.update(geom, b_key_id);
   }
 
   if (!sync) {

intern/cycles/blender/blender_gpu_display.cpp

@@ -0,0 +1,787 @@
+/*
+ * 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.
+ */
+
+#include "blender/blender_gpu_display.h"
+
+#include "device/device.h"
+#include "util/util_logging.h"
+#include "util/util_opengl.h"
+
+extern "C" {
+struct RenderEngine;
+
+bool RE_engine_has_render_context(struct RenderEngine *engine);
+void RE_engine_render_context_enable(struct RenderEngine *engine);
+void RE_engine_render_context_disable(struct RenderEngine *engine);
+
+bool DRW_opengl_context_release();
+void DRW_opengl_context_activate(bool drw_state);
+
+void *WM_opengl_context_create();
+void WM_opengl_context_activate(void *gl_context);
+void WM_opengl_context_dispose(void *gl_context);
+void WM_opengl_context_release(void *context);
+}
+
+CCL_NAMESPACE_BEGIN
+
+/* --------------------------------------------------------------------
+ * BlenderDisplayShader.
+ */
+
+unique_ptr<BlenderDisplayShader> BlenderDisplayShader::create(BL::RenderEngine &b_engine,
+                                                              BL::Scene &b_scene)
+{
+  if (b_engine.support_display_space_shader(b_scene)) {
+    return make_unique<BlenderDisplaySpaceShader>(b_engine, b_scene);
+  }
+
+  return make_unique<BlenderFallbackDisplayShader>();
+}
+
+int BlenderDisplayShader::get_position_attrib_location()
+{
+  if (position_attribute_location_ == -1) {
+    const uint shader_program = get_shader_program();
+    position_attribute_location_ = glGetAttribLocation(shader_program, position_attribute_name);
+  }
+  return position_attribute_location_;
+}
+
+int BlenderDisplayShader::get_tex_coord_attrib_location()
+{
+  if (tex_coord_attribute_location_ == -1) {
+    const uint shader_program = get_shader_program();
+    tex_coord_attribute_location_ = glGetAttribLocation(shader_program, tex_coord_attribute_name);
+  }
+  return tex_coord_attribute_location_;
+}
+
+/* --------------------------------------------------------------------
+ * BlenderFallbackDisplayShader.
+ */
+
+/* TODO move shaders to standalone .glsl file. */
+static const char *FALLBACK_VERTEX_SHADER =
+    "#version 330\n"
+    "uniform vec2 fullscreen;\n"
+    "in vec2 texCoord;\n"
+    "in vec2 pos;\n"
+    "out vec2 texCoord_interp;\n"
+    "\n"
+    "vec2 normalize_coordinates()\n"
+    "{\n"
+    "   return (vec2(2.0) * (pos / fullscreen)) - vec2(1.0);\n"
+    "}\n"
+    "\n"
+    "void main()\n"
+    "{\n"
+    "   gl_Position = vec4(normalize_coordinates(), 0.0, 1.0);\n"
+    "   texCoord_interp = texCoord;\n"
+    "}\n\0";
+
+static const char *FALLBACK_FRAGMENT_SHADER =
+    "#version 330\n"
+    "uniform sampler2D image_texture;\n"
+    "in vec2 texCoord_interp;\n"
+    "out vec4 fragColor;\n"
+    "\n"
+    "void main()\n"
+    "{\n"
+    "   fragColor = texture(image_texture, texCoord_interp);\n"
+    "}\n\0";
+
+static void shader_print_errors(const char *task, const char *log, const char *code)
+{
+  LOG(ERROR) << "Shader: " << task << " error:";
+  LOG(ERROR) << "===== shader string ====";
+
+  stringstream stream(code);
+  string partial;
+
+  int line = 1;
+  while (getline(stream, partial, '\n')) {
+    if (line < 10) {
+      LOG(ERROR) << " " << line << " " << partial;
+    }
+    else {
+      LOG(ERROR) << line << " " << partial;
+    }
+    line++;
+  }
+  LOG(ERROR) << log;
+}
+
+static int compile_fallback_shader(void)
+{
+  const struct Shader {
+    const char *source;
+    const GLenum type;
+  } shaders[2] = {{FALLBACK_VERTEX_SHADER, GL_VERTEX_SHADER},
+                  {FALLBACK_FRAGMENT_SHADER, GL_FRAGMENT_SHADER}};
+
+  const GLuint program = glCreateProgram();
+
+  for (int i = 0; i < 2; i++) {
+    const GLuint shader = glCreateShader(shaders[i].type);
+
+    string source_str = shaders[i].source;
+    const char *c_str = source_str.c_str();
+
+    glShaderSource(shader, 1, &c_str, NULL);
+    glCompileShader(shader);
+
+    GLint compile_status;
+    glGetShaderiv(shader, GL_COMPILE_STATUS, &compile_status);
+
+    if (!compile_status) {
+      GLchar log[5000];
+      GLsizei length = 0;
+      glGetShaderInfoLog(shader, sizeof(log), &length, log);
+      shader_print_errors("compile", log, c_str);
+      return 0;
+    }
+
+    glAttachShader(program, shader);
+  }
+
+  /* Link output. */
+  glBindFragDataLocation(program, 0, "fragColor");
+
+  /* Link and error check. */
+  glLinkProgram(program);
+
+  /* TODO(sergey): Find a way to nicely de-duplicate the error checking. */
+  GLint link_status;
+  glGetProgramiv(program, GL_LINK_STATUS, &link_status);
+  if (!link_status) {
+    GLchar log[5000];
+    GLsizei length = 0;
+    /* TODO(sergey): Is it really program passed to glGetShaderInfoLog? */
+    glGetShaderInfoLog(program, sizeof(log), &length, log);
+    shader_print_errors("linking", log, FALLBACK_VERTEX_SHADER);
+    shader_print_errors("linking", log, FALLBACK_FRAGMENT_SHADER);
+    return 0;
+  }
+
+  return program;
+}
+
+void BlenderFallbackDisplayShader::bind(int width, int height)
+{
+  create_shader_if_needed();
+
+  if (!shader_program_) {
+    return;
+  }
+
+  glUseProgram(shader_program_);
+  glUniform1i(image_texture_location_, 0);
+  glUniform2f(fullscreen_location_, width, height);
+}
+
+void BlenderFallbackDisplayShader::unbind()
+{
+}
+
+uint BlenderFallbackDisplayShader::get_shader_program()
+{
+  return shader_program_;
+}
+
+void BlenderFallbackDisplayShader::create_shader_if_needed()
+{
+  if (shader_program_ || shader_compile_attempted_) {
+    return;
+  }
+
+  shader_compile_attempted_ = true;
+
+  shader_program_ = compile_fallback_shader();
+  if (!shader_program_) {
+    return;
+  }
+
+  glUseProgram(shader_program_);
+
+  image_texture_location_ = glGetUniformLocation(shader_program_, "image_texture");
+  if (image_texture_location_ < 0) {
+    LOG(ERROR) << "Shader doesn't contain the 'image_texture' uniform.";
+    destroy_shader();
+    return;
+  }
+
+  fullscreen_location_ = glGetUniformLocation(shader_program_, "fullscreen");
+  if (fullscreen_location_ < 0) {
+    LOG(ERROR) << "Shader doesn't contain the 'fullscreen' uniform.";
+    destroy_shader();
+    return;
+  }
+}
+
+void BlenderFallbackDisplayShader::destroy_shader()
+{
+  glDeleteProgram(shader_program_);
+  shader_program_ = 0;
+}
+
+/* --------------------------------------------------------------------
+ * BlenderDisplaySpaceShader.
+ */
+
+BlenderDisplaySpaceShader::BlenderDisplaySpaceShader(BL::RenderEngine &b_engine,
+                                                     BL::Scene &b_scene)
+    : b_engine_(b_engine), b_scene_(b_scene)
+{
+  DCHECK(b_engine_.support_display_space_shader(b_scene_));
+}
+
+void BlenderDisplaySpaceShader::bind(int /*width*/, int /*height*/)
+{
+  b_engine_.bind_display_space_shader(b_scene_);
+}
+
+void BlenderDisplaySpaceShader::unbind()
+{
+  b_engine_.unbind_display_space_shader();
+}
+
+uint BlenderDisplaySpaceShader::get_shader_program()
+{
+  if (!shader_program_) {
+    glGetIntegerv(GL_CURRENT_PROGRAM, reinterpret_cast<int *>(&shader_program_));
+  }
+
+  if (!shader_program_) {
+    LOG(ERROR) << "Error retrieving shader program for display space shader.";
+  }
+
+  return shader_program_;
+}
+
+/* --------------------------------------------------------------------
+ * BlenderGPUDisplay.
+ */
+
+BlenderGPUDisplay::BlenderGPUDisplay(BL::RenderEngine &b_engine, BL::Scene &b_scene)
+    : b_engine_(b_engine), display_shader_(BlenderDisplayShader::create(b_engine, b_scene))
+{
+  /* Create context while on the main thread. */
+  gl_context_create();
+}
+
+BlenderGPUDisplay::~BlenderGPUDisplay()
+{
+  gl_resources_destroy();
+}
+
+/* --------------------------------------------------------------------
+ * Update procedure.
+ */
+
+bool BlenderGPUDisplay::do_update_begin(const GPUDisplayParams &params,
+                                        int texture_width,
+                                        int texture_height)
+{
+  /* Note that it's the responsibility of BlenderGPUDisplay to ensure updating and drawing
+   * the texture does not happen at the same time. This is achieved indirectly.
+   *
+   * When enabling the OpenGL context, it uses an internal mutex lock DST.gl_context_lock.
+   * This same lock is also held when do_draw() is called, which together ensure mutual
+   * exclusion.
+   *
+   * This locking is not performed at the GPU display level, because that would cause lock
+   * inversion. */
+  if (!gl_context_enable()) {
+    return false;
+  }
+
+  if (gl_render_sync_) {
+    glWaitSync((GLsync)gl_render_sync_, 0, GL_TIMEOUT_IGNORED);
+  }
+
+  if (!gl_texture_resources_ensure()) {
+    gl_context_disable();
+    return false;
+  }
+
+  /* Update texture dimensions if needed. */
+  if (texture_.width != texture_width || texture_.height != texture_height) {
+    glActiveTexture(GL_TEXTURE0);
+    glBindTexture(GL_TEXTURE_2D, texture_.gl_id);
+    glTexImage2D(
+        GL_TEXTURE_2D, 0, GL_RGBA16F, texture_width, texture_height, 0, GL_RGBA, GL_HALF_FLOAT, 0);
+    texture_.width = texture_width;
+    texture_.height = texture_height;
+    glBindTexture(GL_TEXTURE_2D, 0);
+
+    /* Texture did change, and no pixel storage was provided. Tag for an explicit zeroing out to
+     * avoid undefined content. */
+    texture_.need_clear = true;
+  }
+
+  /* Update PBO dimensions if needed.
+   *
+   * NOTE: Allocate the PBO for the the size which will fit the final render resolution (as in,
+   * at a resolution divider 1. This was we don't need to recreate graphics interoperability
+   * objects which are costly and which are tied to the specific underlying buffer size.
+   * The downside of this approach is that when graphics interoperability is not used we are
+   * sending too much data to GPU when resolution divider is not 1. */
+  /* TODO(sergey): Investigate whether keeping the PBO exact size of the texture makes non-interop
+   * mode faster. */
+  const int buffer_width = params.full_size.x;
+  const int buffer_height = params.full_size.y;
+  if (texture_.buffer_width != buffer_width || texture_.buffer_height != buffer_height) {
+    const size_t size_in_bytes = sizeof(half4) * buffer_width * buffer_height;
+    glBindBuffer(GL_PIXEL_UNPACK_BUFFER, texture_.gl_pbo_id);
+    glBufferData(GL_PIXEL_UNPACK_BUFFER, size_in_bytes, 0, GL_DYNAMIC_DRAW);
+    glBindBuffer(GL_PIXEL_UNPACK_BUFFER, 0);
+
+    texture_.buffer_width = buffer_width;
+    texture_.buffer_height = buffer_height;
+  }
+
+  /* New content will be provided to the texture in one way or another, so mark this in a
+   * centralized place. */
+  texture_.need_update = true;
+
+  return true;
+}
+
+void BlenderGPUDisplay::do_update_end()
+{
+  gl_upload_sync_ = glFenceSync(GL_SYNC_GPU_COMMANDS_COMPLETE, 0);
+  glFlush();
+
+  gl_context_disable();
+}
+
+/* --------------------------------------------------------------------
+ * Texture update from CPU buffer.
+ */
+
+void BlenderGPUDisplay::do_copy_pixels_to_texture(
+    const half4 *rgba_pixels, int texture_x, int texture_y, int pixels_width, int pixels_height)
+{
+  /* This call copies pixels to a Pixel Buffer Object (PBO) which is much cheaper from CPU time
+   * point of view than to copy data directly to the OpenGL texture.
+   *
+   * The possible downside of this approach is that it might require a higher peak memory when
+   * doing partial updates of the texture (although, in practice even partial updates might peak
+   * with a full-frame buffer stored on the CPU if the GPU is currently occupied). */
+
+  half4 *mapped_rgba_pixels = map_texture_buffer();
+  if (!mapped_rgba_pixels) {
+    return;
+  }
+
+  if (texture_x == 0 && texture_y == 0 && pixels_width == texture_.width &&
+      pixels_height == texture_.height) {
+    const size_t size_in_bytes = sizeof(half4) * texture_.width * texture_.height;
+    memcpy(mapped_rgba_pixels, rgba_pixels, size_in_bytes);
+  }
+  else {
+    const half4 *rgba_row = rgba_pixels;
+    half4 *mapped_rgba_row = mapped_rgba_pixels + texture_y * texture_.width + texture_x;
+    for (int y = 0; y < pixels_height;
+         ++y, rgba_row += pixels_width, mapped_rgba_row += texture_.width) {
+      memcpy(mapped_rgba_row, rgba_row, sizeof(half4) * pixels_width);
+    }
+  }
+
+  unmap_texture_buffer();
+}
+
+/* --------------------------------------------------------------------
+ * Texture buffer mapping.
+ */
+
+half4 *BlenderGPUDisplay::do_map_texture_buffer()
+{
+  glBindBuffer(GL_PIXEL_UNPACK_BUFFER, texture_.gl_pbo_id);
+
+  half4 *mapped_rgba_pixels = reinterpret_cast<half4 *>(
+      glMapBuffer(GL_PIXEL_UNPACK_BUFFER, GL_WRITE_ONLY));
+  if (!mapped_rgba_pixels) {
+    LOG(ERROR) << "Error mapping BlenderGPUDisplay pixel buffer object.";
+  }
+
+  if (texture_.need_clear) {
+    const int64_t texture_width = texture_.width;
+    const int64_t texture_height = texture_.height;
+    memset(reinterpret_cast<void *>(mapped_rgba_pixels),
+           0,
+           texture_width * texture_height * sizeof(half4));
+    texture_.need_clear = false;
+  }
+
+  return mapped_rgba_pixels;
+}
+
+void BlenderGPUDisplay::do_unmap_texture_buffer()
+{
+  glUnmapBuffer(GL_PIXEL_UNPACK_BUFFER);
+
+  glBindBuffer(GL_PIXEL_UNPACK_BUFFER, 0);
+}
+
+/* --------------------------------------------------------------------
+ * Graphics interoperability.
+ */
+
+DeviceGraphicsInteropDestination BlenderGPUDisplay::do_graphics_interop_get()
+{
+  DeviceGraphicsInteropDestination interop_dst;
+
+  interop_dst.buffer_width = texture_.buffer_width;
+  interop_dst.buffer_height = texture_.buffer_height;
+  interop_dst.opengl_pbo_id = texture_.gl_pbo_id;
+
+  interop_dst.need_clear = texture_.need_clear;
+  texture_.need_clear = false;
+
+  return interop_dst;
+}
+
+void BlenderGPUDisplay::graphics_interop_activate()
+{
+  gl_context_enable();
+}
+
+void BlenderGPUDisplay::graphics_interop_deactivate()
+{
+  gl_context_disable();
+}
+
+/* --------------------------------------------------------------------
+ * Drawing.
+ */
+
+void BlenderGPUDisplay::clear()
+{
+  texture_.need_clear = true;
+}
+
+void BlenderGPUDisplay::set_zoom(float zoom_x, float zoom_y)
+{
+  zoom_ = make_float2(zoom_x, zoom_y);
+}
+
+void BlenderGPUDisplay::do_draw(const GPUDisplayParams &params)
+{
+  /* See do_update_begin() for why no locking is required here. */
+  const bool transparent = true;  // TODO(sergey): Derive this from Film.
+
+  if (texture_.need_clear) {
+    /* Texture is requested to be cleared and was not yet cleared.
+     * Do early return which should be equivalent of drawing all-zero texture. */
+    return;
+  }
+
+  if (!gl_draw_resources_ensure()) {
+    return;
+  }
+
+  if (use_gl_context_) {
+    gl_context_mutex_.lock();
+  }
+
+  if (gl_upload_sync_) {
+    glWaitSync((GLsync)gl_upload_sync_, 0, GL_TIMEOUT_IGNORED);
+  }
+
+  if (transparent) {
+    glEnable(GL_BLEND);
+    glBlendFunc(GL_ONE, GL_ONE_MINUS_SRC_ALPHA);
+  }
+
+  display_shader_->bind(params.full_size.x, params.full_size.y);
+
+  glActiveTexture(GL_TEXTURE0);
+  glBindTexture(GL_TEXTURE_2D, texture_.gl_id);
+
+  /* Trick to keep sharp rendering without jagged edges on all GPUs.
+   *
+   * The idea here is to enforce driver to use linear interpolation when the image is not zoomed
+   * in.
+   * For the render result with a resolution divider in effect we always use nearest interpolation.
+   *
+   * Use explicit MIN assignment to make sure the driver does not have an undefined behavior at
+   * the zoom level 1. The MAG filter is always NEAREST. */
+  const float zoomed_width = params.size.x * zoom_.x;
+  const float zoomed_height = params.size.y * zoom_.y;
+  if (texture_.width != params.size.x || texture_.height != params.size.y) {
+    /* Resolution divider is different from 1, force nearest interpolation. */
+    glTexParameteri(GL_TEXTURE_2D, GL_TEXTURE_MIN_FILTER, GL_NEAREST);
+  }
+  else if (zoomed_width - params.size.x > 0.5f || zoomed_height - params.size.y > 0.5f) {
+    glTexParameteri(GL_TEXTURE_2D, GL_TEXTURE_MIN_FILTER, GL_NEAREST);
+  }
+  else {
+    glTexParameteri(GL_TEXTURE_2D, GL_TEXTURE_MIN_FILTER, GL_LINEAR);
+  }
+
+  glBindBuffer(GL_ARRAY_BUFFER, vertex_buffer_);
+
+  texture_update_if_needed();
+  vertex_buffer_update(params);
+
+  /* TODO(sergey): Does it make sense/possible to cache/reuse the VAO? */
+  GLuint vertex_array_object;
+  glGenVertexArrays(1, &vertex_array_object);
+  glBindVertexArray(vertex_array_object);
+
+  const int texcoord_attribute = display_shader_->get_tex_coord_attrib_location();
+  const int position_attribute = display_shader_->get_position_attrib_location();
+
+  glEnableVertexAttribArray(texcoord_attribute);
+  glEnableVertexAttribArray(position_attribute);
+
+  glVertexAttribPointer(
+      texcoord_attribute, 2, GL_FLOAT, GL_FALSE, 4 * sizeof(float), (const GLvoid *)0);
+  glVertexAttribPointer(position_attribute,
+                        2,
+                        GL_FLOAT,
+                        GL_FALSE,
+                        4 * sizeof(float),
+                        (const GLvoid *)(sizeof(float) * 2));
+
+  glDrawArrays(GL_TRIANGLE_FAN, 0, 4);
+
+  glBindBuffer(GL_ARRAY_BUFFER, 0);
+  glBindTexture(GL_TEXTURE_2D, 0);
+
+  glDeleteVertexArrays(1, &vertex_array_object);
+
+  display_shader_->unbind();
+
+  if (transparent) {
+    glDisable(GL_BLEND);
+  }
+
+  gl_render_sync_ = glFenceSync(GL_SYNC_GPU_COMMANDS_COMPLETE, 0);
+  glFlush();
+
+  if (use_gl_context_) {
+    gl_context_mutex_.unlock();
+  }
+}
+
+void BlenderGPUDisplay::gl_context_create()
+{
+  /* When rendering in viewport there is no render context available via engine.
+   * Check whether own context is to be created here.
+   *
+   * NOTE: If the `b_engine_`'s context is not available, we are expected to be on a main thread
+   * here. */
+  use_gl_context_ = !RE_engine_has_render_context(
+      reinterpret_cast<RenderEngine *>(b_engine_.ptr.data));
+
+  if (use_gl_context_) {
+    const bool drw_state = DRW_opengl_context_release();
+    gl_context_ = WM_opengl_context_create();
+    if (gl_context_) {
+      /* On Windows an old context is restored after creation, and subsequent release of context
+       * generates a Win32 error. Harmless for users, but annoying to have possible misleading
+       * error prints in the console. */
+#ifndef _WIN32
+      WM_opengl_context_release(gl_context_);
+#endif
+    }
+    else {
+      LOG(ERROR) << "Error creating OpenGL context.";
+    }
+
+    DRW_opengl_context_activate(drw_state);
+  }
+}
+
+bool BlenderGPUDisplay::gl_context_enable()
+{
+  if (use_gl_context_) {
+    if (!gl_context_) {
+      return false;
+    }
+    gl_context_mutex_.lock();
+    WM_opengl_context_activate(gl_context_);
+    return true;
+  }
+
+  RE_engine_render_context_enable(reinterpret_cast<RenderEngine *>(b_engine_.ptr.data));
+  return true;
+}
+
+void BlenderGPUDisplay::gl_context_disable()
+{
+  if (use_gl_context_) {
+    if (gl_context_) {
+      WM_opengl_context_release(gl_context_);
+      gl_context_mutex_.unlock();
+    }
+    return;
+  }
+
+  RE_engine_render_context_disable(reinterpret_cast<RenderEngine *>(b_engine_.ptr.data));
+}
+
+void BlenderGPUDisplay::gl_context_dispose()
+{
+  if (gl_context_) {
+    const bool drw_state = DRW_opengl_context_release();
+
+    WM_opengl_context_activate(gl_context_);
+    WM_opengl_context_dispose(gl_context_);
+
+    DRW_opengl_context_activate(drw_state);
+  }
+}
+
+bool BlenderGPUDisplay::gl_draw_resources_ensure()
+{
+  if (!texture_.gl_id) {
+    /* If there is no texture allocated, there is nothing to draw. Inform the draw call that it can
+     * can not continue. Note that this is not an unrecoverable error, so once the texture is known
+     * we will come back here and create all the GPU resources needed for draw. */
+    return false;
+  }
+
+  if (gl_draw_resource_creation_attempted_) {
+    return gl_draw_resources_created_;
+  }
+  gl_draw_resource_creation_attempted_ = true;
+
+  if (!vertex_buffer_) {
+    glGenBuffers(1, &vertex_buffer_);
+    if (!vertex_buffer_) {
+      LOG(ERROR) << "Error creating vertex buffer.";
+      return false;
+    }
+  }
+
+  gl_draw_resources_created_ = true;
+
+  return true;
+}
+
+void BlenderGPUDisplay::gl_resources_destroy()
+{
+  gl_context_enable();
+
+  if (vertex_buffer_ != 0) {
+    glDeleteBuffers(1, &vertex_buffer_);
+  }
+
+  if (texture_.gl_pbo_id) {
+    glDeleteBuffers(1, &texture_.gl_pbo_id);
+    texture_.gl_pbo_id = 0;
+  }
+
+  if (texture_.gl_id) {
+    glDeleteTextures(1, &texture_.gl_id);
+    texture_.gl_id = 0;
+  }
+
+  gl_context_disable();
+
+  gl_context_dispose();
+}
+
+bool BlenderGPUDisplay::gl_texture_resources_ensure()
+{
+  if (texture_.creation_attempted) {
+    return texture_.is_created;
+  }
+  texture_.creation_attempted = true;
+
+  DCHECK(!texture_.gl_id);
+  DCHECK(!texture_.gl_pbo_id);
+
+  /* Create texture. */
+  glGenTextures(1, &texture_.gl_id);
+  if (!texture_.gl_id) {
+    LOG(ERROR) << "Error creating texture.";
+    return false;
+  }
+
+  /* Configure the texture. */
+  glActiveTexture(GL_TEXTURE0);
+  glBindTexture(GL_TEXTURE_2D, texture_.gl_id);
+  glTexParameteri(GL_TEXTURE_2D, GL_TEXTURE_MIN_FILTER, GL_NEAREST);
+  glTexParameteri(GL_TEXTURE_2D, GL_TEXTURE_MAG_FILTER, GL_NEAREST);
+  glBindTexture(GL_TEXTURE_2D, 0);
+
+  /* Create PBO for the texture. */
+  glGenBuffers(1, &texture_.gl_pbo_id);
+  if (!texture_.gl_pbo_id) {
+    LOG(ERROR) << "Error creating texture pixel buffer object.";
+    return false;
+  }
+
+  /* Creation finished with a success. */
+  texture_.is_created = true;
+
+  return true;
+}
+
+void BlenderGPUDisplay::texture_update_if_needed()
+{
+  if (!texture_.need_update) {
+    return;
+  }
+
+  glBindBuffer(GL_PIXEL_UNPACK_BUFFER, texture_.gl_pbo_id);
+  glTexSubImage2D(
+      GL_TEXTURE_2D, 0, 0, 0, texture_.width, texture_.height, GL_RGBA, GL_HALF_FLOAT, 0);
+  glBindBuffer(GL_PIXEL_UNPACK_BUFFER, 0);
+
+  texture_.need_update = false;
+}
+
+void BlenderGPUDisplay::vertex_buffer_update(const GPUDisplayParams &params)
+{
+  /* Invalidate old contents - avoids stalling if the buffer is still waiting in queue to be
+   * rendered. */
+  glBufferData(GL_ARRAY_BUFFER, 16 * sizeof(float), NULL, GL_STREAM_DRAW);
+
+  float *vpointer = reinterpret_cast<float *>(glMapBuffer(GL_ARRAY_BUFFER, GL_WRITE_ONLY));
+  if (!vpointer) {
+    return;
+  }
+
+  vpointer[0] = 0.0f;
+  vpointer[1] = 0.0f;
+  vpointer[2] = params.offset.x;
+  vpointer[3] = params.offset.y;
+
+  vpointer[4] = 1.0f;
+  vpointer[5] = 0.0f;
+  vpointer[6] = (float)params.size.x + params.offset.x;
+  vpointer[7] = params.offset.y;
+
+  vpointer[8] = 1.0f;
+  vpointer[9] = 1.0f;
+  vpointer[10] = (float)params.size.x + params.offset.x;
+  vpointer[11] = (float)params.size.y + params.offset.y;
+
+  vpointer[12] = 0.0f;
+  vpointer[13] = 1.0f;
+  vpointer[14] = params.offset.x;
+  vpointer[15] = (float)params.size.y + params.offset.y;
+
+  glUnmapBuffer(GL_ARRAY_BUFFER);
+}
+
+CCL_NAMESPACE_END

intern/cycles/blender/blender_gpu_display.h

@@ -0,0 +1,215 @@
+/*
+ * 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
+
+#include <atomic>
+
+#include "MEM_guardedalloc.h"
+
+#include "RNA_blender_cpp.h"
+
+#include "render/gpu_display.h"
+#include "util/util_unique_ptr.h"
+
+CCL_NAMESPACE_BEGIN
+
+/* Base class of shader used for GPU display rendering. */
+class BlenderDisplayShader {
+ public:
+  static constexpr const char *position_attribute_name = "pos";
+  static constexpr const char *tex_coord_attribute_name = "texCoord";
+
+  /* Create shader implementation suitable for the given render engine and scene configuration. */
+  static unique_ptr<BlenderDisplayShader> create(BL::RenderEngine &b_engine, BL::Scene &b_scene);
+
+  BlenderDisplayShader() = default;
+  virtual ~BlenderDisplayShader() = default;
+
+  virtual void bind(int width, int height) = 0;
+  virtual void unbind() = 0;
+
+  /* Get attribute location for position and texture coordinate respectively.
+   * NOTE: The shader needs to be bound to have access to those. */
+  virtual int get_position_attrib_location();
+  virtual int get_tex_coord_attrib_location();
+
+ protected:
+  /* Get program of this display shader.
+   * NOTE: The shader needs to be bound to have access to this. */
+  virtual uint get_shader_program() = 0;
+
+  /* Cached values of various OpenGL resources. */
+  int position_attribute_location_ = -1;
+  int tex_coord_attribute_location_ = -1;
+};
+
+/* Implementation of display rendering shader used in the case when render engine does not support
+ * display space shader. */
+class BlenderFallbackDisplayShader : public BlenderDisplayShader {
+ public:
+  virtual void bind(int width, int height) override;
+  virtual void unbind() override;
+
+ protected:
+  virtual uint get_shader_program() override;
+
+  void create_shader_if_needed();
+  void destroy_shader();
+
+  uint shader_program_ = 0;
+  int image_texture_location_ = -1;
+  int fullscreen_location_ = -1;
+
+  /* Shader compilation attempted. Which means, that if the shader program is 0 then compilation or
+   * linking has failed. Do not attempt to re-compile the shader. */
+  bool shader_compile_attempted_ = false;
+};
+
+class BlenderDisplaySpaceShader : public BlenderDisplayShader {
+ public:
+  BlenderDisplaySpaceShader(BL::RenderEngine &b_engine, BL::Scene &b_scene);
+
+  virtual void bind(int width, int height) override;
+  virtual void unbind() override;
+
+ protected:
+  virtual uint get_shader_program() override;
+
+  BL::RenderEngine b_engine_;
+  BL::Scene &b_scene_;
+
+  /* Cached values of various OpenGL resources. */
+  uint shader_program_ = 0;
+};
+
+/* GPU display implementation which is specific for Blender viewport integration. */
+class BlenderGPUDisplay : public GPUDisplay {
+ public:
+  BlenderGPUDisplay(BL::RenderEngine &b_engine, BL::Scene &b_scene);
+  ~BlenderGPUDisplay();
+
+  virtual void graphics_interop_activate() override;
+  virtual void graphics_interop_deactivate() override;
+
+  virtual void clear() override;
+
+  void set_zoom(float zoom_x, float zoom_y);
+
+ protected:
+  virtual bool do_update_begin(const GPUDisplayParams &params,
+                               int texture_width,
+                               int texture_height) override;
+  virtual void do_update_end() override;
+
+  virtual void do_copy_pixels_to_texture(const half4 *rgba_pixels,
+                                         int texture_x,
+                                         int texture_y,
+                                         int pixels_width,
+                                         int pixels_height) override;
+  virtual void do_draw(const GPUDisplayParams &params) override;
+
+  virtual half4 *do_map_texture_buffer() override;
+  virtual void do_unmap_texture_buffer() override;
+
+  virtual DeviceGraphicsInteropDestination do_graphics_interop_get() override;
+
+  /* Helper function which allocates new GPU context. */
+  void gl_context_create();
+  bool gl_context_enable();
+  void gl_context_disable();
+  void gl_context_dispose();
+
+  /* Make sure texture is allocated and its initial configuration is performed. */
+  bool gl_texture_resources_ensure();
+
+  /* Ensure all runtime GPU resources needed for drawing are allocated.
+   * Returns true if all resources needed for drawing are available. */
+  bool gl_draw_resources_ensure();
+
+  /* Destroy all GPU resources which are being used by this object. */
+  void gl_resources_destroy();
+
+  /* Update GPU texture dimensions and content if needed (new pixel data was provided).
+   *
+   * NOTE: The texture needs to be bound. */
+  void texture_update_if_needed();
+
+  /* Update vertex buffer with new coordinates of vertex positions and texture coordinates.
+   * This buffer is used to render texture in the viewport.
+   *
+   * NOTE: The buffer needs to be bound. */
+  void vertex_buffer_update(const GPUDisplayParams &params);
+
+  BL::RenderEngine b_engine_;
+
+  /* OpenGL context which is used the render engine doesn't have its own. */
+  void *gl_context_ = nullptr;
+  /* The when Blender RenderEngine side context is not available and the GPUDisplay is to create
+   * its own context. */
+  bool use_gl_context_ = false;
+  /* Mutex used to guard the `gl_context_`. */
+  thread_mutex gl_context_mutex_;
+
+  /* Texture which contains pixels of the render result. */
+  struct {
+    /* Indicates whether texture creation was attempted and succeeded.
+     * Used to avoid multiple attempts of texture creation on GPU issues or GPU context
+     * misconfiguration. */
+    bool creation_attempted = false;
+    bool is_created = false;
+
+    /* OpenGL resource IDs of the texture itself and Pixel Buffer Object (PBO) used to write
+     * pixels to it.
+     *
+     * NOTE: Allocated on the engine's context. */
+    uint gl_id = 0;
+    uint gl_pbo_id = 0;
+
+    /* Is true when new data was written to the PBO, meaning, the texture might need to be resized
+     * and new data is to be uploaded to the GPU. */
+    bool need_update = false;
+
+    /* Content of the texture is to be filled with zeroes. */
+    std::atomic<bool> need_clear = true;
+
+    /* Dimensions of the texture in pixels. */
+    int width = 0;
+    int height = 0;
+
+    /* Dimensions of the underlying PBO. */
+    int buffer_width = 0;
+    int buffer_height = 0;
+  } texture_;
+
+  unique_ptr<BlenderDisplayShader> display_shader_;
+
+  /* Special track of whether GPU resources were attempted to be created, to avoid attempts of
+   * their re-creation on failure on every redraw. */
+  bool gl_draw_resource_creation_attempted_ = false;
+  bool gl_draw_resources_created_ = false;
+
+  /* Vertex buffer which hold vertices of a triangle fan which is textures with the texture
+   * holding the render result. */
+  uint vertex_buffer_ = 0;
+
+  void *gl_render_sync_ = nullptr;
+  void *gl_upload_sync_ = nullptr;
+
+  float2 zoom_ = make_float2(1.0f, 1.0f);
+};
+
+CCL_NAMESPACE_END

intern/cycles/blender/blender_light.cpp

@@ -125,17 +125,10 @@ void BlenderSync::sync_light(BL::Object &b_parent,
   light->set_shader(static_cast<Shader *>(used_shaders[0]));
 
   /* shadow */
-  PointerRNA cscene = RNA_pointer_get(&b_scene.ptr, "cycles");
   PointerRNA clight = RNA_pointer_get(&b_light.ptr, "cycles");
   light->set_cast_shadow(get_boolean(clight, "cast_shadow"));
   light->set_use_mis(get_boolean(clight, "use_multiple_importance_sampling"));
 
-  int samples = get_int(clight, "samples");
-  if (get_boolean(cscene, "use_square_samples"))
-    light->set_samples(samples * samples);
-  else
-    light->set_samples(samples);
-
   light->set_max_bounces(get_int(clight, "max_bounces"));
 
   if (b_ob_info.real_object != b_ob_info.iter_object) {
@@ -155,10 +148,12 @@ void BlenderSync::sync_light(BL::Object &b_parent,
 
   /* visibility */
   uint visibility = object_ray_visibility(b_ob_info.real_object);
+  light->set_use_camera((visibility & PATH_RAY_CAMERA) != 0);
   light->set_use_diffuse((visibility & PATH_RAY_DIFFUSE) != 0);
   light->set_use_glossy((visibility & PATH_RAY_GLOSSY) != 0);
   light->set_use_transmission((visibility & PATH_RAY_TRANSMIT) != 0);
   light->set_use_scatter((visibility & PATH_RAY_VOLUME_SCATTER) != 0);
+  light->set_is_shadow_catcher(b_ob_info.real_object.is_shadow_catcher());
 
   /* tag */
   light->tag_update(scene);
@@ -169,7 +164,6 @@ void BlenderSync::sync_background_light(BL::SpaceView3D &b_v3d, bool use_portal)
   BL::World b_world = b_scene.world();
 
   if (b_world) {
-    PointerRNA cscene = RNA_pointer_get(&b_scene.ptr, "cycles");
     PointerRNA cworld = RNA_pointer_get(&b_world.ptr, "cycles");
 
     enum SamplingMethod { SAMPLING_NONE = 0, SAMPLING_AUTOMATIC, SAMPLING_MANUAL, SAMPLING_NUM };
@@ -197,12 +191,6 @@ void BlenderSync::sync_background_light(BL::SpaceView3D &b_v3d, bool use_portal)
         /* force enable light again when world is resynced */
         light->set_is_enabled(true);
 
-        int samples = get_int(cworld, "samples");
-        if (get_boolean(cscene, "use_square_samples"))
-          light->set_samples(samples * samples);
-        else
-          light->set_samples(samples);
-
         light->tag_update(scene);
         light_map.set_recalc(b_world);
       }
@@ -211,7 +199,7 @@ void BlenderSync::sync_background_light(BL::SpaceView3D &b_v3d, bool use_portal)
 
   world_map = b_world.ptr.data;
   world_recalc = false;
-  viewport_parameters = BlenderViewportParameters(b_v3d);
+  viewport_parameters = BlenderViewportParameters(b_v3d, use_developer_ui);
 }
 
 CCL_NAMESPACE_END

intern/cycles/blender/blender_object.cpp

@@ -568,7 +568,7 @@ void BlenderSync::sync_objects(BL::Depsgraph &b_depsgraph,
   /* object loop */
   bool cancel = false;
   bool use_portal = false;
-  const bool show_lights = BlenderViewportParameters(b_v3d).use_scene_lights;
+  const bool show_lights = BlenderViewportParameters(b_v3d, use_developer_ui).use_scene_lights;
 
   BL::ViewLayer b_view_layer = b_depsgraph.view_layer_eval();
   BL::Depsgraph::object_instances_iterator b_instance_iter;

intern/cycles/blender/blender_python.cpp

@@ -45,10 +45,6 @@
 #  include <OSL/oslquery.h>
 #endif
 
-#ifdef WITH_OPENCL
-#  include "device/device_intern.h"
-#endif
-
 CCL_NAMESPACE_BEGIN
 
 namespace {
@@ -72,12 +68,10 @@ PyObject *pyunicode_from_string(const char *str)
 /* Synchronize debug flags from a given Blender scene.
  * Return truth when device list needs invalidation.
  */
-bool debug_flags_sync_from_scene(BL::Scene b_scene)
+static void debug_flags_sync_from_scene(BL::Scene b_scene)
 {
   DebugFlagsRef flags = DebugFlags();
   PointerRNA cscene = RNA_pointer_get(&b_scene.ptr, "cycles");
-  /* Backup some settings for comparison. */
-  DebugFlags::OpenCL::DeviceType opencl_device_type = flags.opencl.device_type;
   /* Synchronize shared flags. */
   flags.viewport_static_bvh = get_enum(cscene, "debug_bvh_type");
   /* Synchronize CPU flags. */
@@ -87,50 +81,19 @@ bool debug_flags_sync_from_scene(BL::Scene b_scene)
   flags.cpu.sse3 = get_boolean(cscene, "debug_use_cpu_sse3");
   flags.cpu.sse2 = get_boolean(cscene, "debug_use_cpu_sse2");
   flags.cpu.bvh_layout = (BVHLayout)get_enum(cscene, "debug_bvh_layout");
-  flags.cpu.split_kernel = get_boolean(cscene, "debug_use_cpu_split_kernel");
   /* Synchronize CUDA flags. */
   flags.cuda.adaptive_compile = get_boolean(cscene, "debug_use_cuda_adaptive_compile");
-  flags.cuda.split_kernel = get_boolean(cscene, "debug_use_cuda_split_kernel");
   /* Synchronize OptiX flags. */
-  flags.optix.cuda_streams = get_int(cscene, "debug_optix_cuda_streams");
-  flags.optix.curves_api = get_boolean(cscene, "debug_optix_curves_api");
-  /* Synchronize OpenCL device type. */
-  switch (get_enum(cscene, "debug_opencl_device_type")) {
-    case 0:
-      flags.opencl.device_type = DebugFlags::OpenCL::DEVICE_NONE;
-      break;
-    case 1:
-      flags.opencl.device_type = DebugFlags::OpenCL::DEVICE_ALL;
-      break;
-    case 2:
-      flags.opencl.device_type = DebugFlags::OpenCL::DEVICE_DEFAULT;
-      break;
-    case 3:
-      flags.opencl.device_type = DebugFlags::OpenCL::DEVICE_CPU;
-      break;
-    case 4:
-      flags.opencl.device_type = DebugFlags::OpenCL::DEVICE_GPU;
-      break;
-    case 5:
-      flags.opencl.device_type = DebugFlags::OpenCL::DEVICE_ACCELERATOR;
-      break;
-  }
-  /* Synchronize other OpenCL flags. */
-  flags.opencl.debug = get_boolean(cscene, "debug_use_opencl_debug");
-  flags.opencl.mem_limit = ((size_t)get_int(cscene, "debug_opencl_mem_limit")) * 1024 * 1024;
-  return flags.opencl.device_type != opencl_device_type;
+  flags.optix.use_debug = get_boolean(cscene, "debug_use_optix_debug");
 }
 
 /* Reset debug flags to default values.
  * Return truth when device list needs invalidation.
  */
-bool debug_flags_reset()
+static void debug_flags_reset()
 {
   DebugFlagsRef flags = DebugFlags();
-  /* Backup some settings for comparison. */
-  DebugFlags::OpenCL::DeviceType opencl_device_type = flags.opencl.device_type;
   flags.reset();
-  return flags.opencl.device_type != opencl_device_type;
 }
 
 } /* namespace */
@@ -175,18 +138,20 @@ static const char *PyC_UnicodeAsByte(PyObject *py_str, PyObject **coerce)
 
 static PyObject *init_func(PyObject * /*self*/, PyObject *args)
 {
-  PyObject *path, *user_path;
+  PyObject *path, *user_path, *temp_path;
   int headless;
 
-  if (!PyArg_ParseTuple(args, "OOi", &path, &user_path, &headless)) {
-    return NULL;
+  if (!PyArg_ParseTuple(args, "OOOi", &path, &user_path, &temp_path, &headless)) {
+    return nullptr;
   }
 
-  PyObject *path_coerce = NULL, *user_path_coerce = NULL;
+  PyObject *path_coerce = nullptr, *user_path_coerce = nullptr, *temp_path_coerce = nullptr;
   path_init(PyC_UnicodeAsByte(path, &path_coerce),
-            PyC_UnicodeAsByte(user_path, &user_path_coerce));
+            PyC_UnicodeAsByte(user_path, &user_path_coerce),
+            PyC_UnicodeAsByte(temp_path, &temp_path_coerce));
   Py_XDECREF(path_coerce);
   Py_XDECREF(user_path_coerce);
+  Py_XDECREF(temp_path_coerce);
 
   BlenderSession::headless = headless;
 
@@ -299,6 +264,50 @@ static PyObject *render_func(PyObject * /*self*/, PyObject *args)
   Py_RETURN_NONE;
 }
 
+static PyObject *render_frame_finish_func(PyObject * /*self*/, PyObject *args)
+{
+  PyObject *pysession;
+
+  if (!PyArg_ParseTuple(args, "O", &pysession)) {
+    return nullptr;
+  }
+
+  BlenderSession *session = (BlenderSession *)PyLong_AsVoidPtr(pysession);
+
+  /* Allow Blender to execute other Python scripts. */
+  python_thread_state_save(&session->python_thread_state);
+
+  session->render_frame_finish();
+
+  python_thread_state_restore(&session->python_thread_state);
+
+  Py_RETURN_NONE;
+}
+
+static PyObject *draw_func(PyObject * /*self*/, PyObject *args)
+{
+  PyObject *py_session, *py_graph, *py_screen, *py_space_image;
+
+  if (!PyArg_ParseTuple(args, "OOOO", &py_session, &py_graph, &py_screen, &py_space_image)) {
+    return nullptr;
+  }
+
+  BlenderSession *session = (BlenderSession *)PyLong_AsVoidPtr(py_session);
+
+  ID *b_screen = (ID *)PyLong_AsVoidPtr(py_screen);
+
+  PointerRNA b_space_image_ptr;
+  RNA_pointer_create(b_screen,
+                     &RNA_SpaceImageEditor,
+                     pylong_as_voidptr_typesafe(py_space_image),
+                     &b_space_image_ptr);
+  BL::SpaceImageEditor b_space_image(b_space_image_ptr);
+
+  session->draw(b_space_image);
+
+  Py_RETURN_NONE;
+}
+
 /* pixel_array and result passed as pointers */
 static PyObject *bake_func(PyObject * /*self*/, PyObject *args)
 {
@@ -336,7 +345,7 @@ static PyObject *bake_func(PyObject * /*self*/, PyObject *args)
   Py_RETURN_NONE;
 }
 
-static PyObject *draw_func(PyObject * /*self*/, PyObject *args)
+static PyObject *view_draw_func(PyObject * /*self*/, PyObject *args)
 {
   PyObject *pysession, *pygraph, *pyv3d, *pyrv3d;
 
@@ -350,7 +359,7 @@ static PyObject *draw_func(PyObject * /*self*/, PyObject *args)
     int viewport[4];
     glGetIntegerv(GL_VIEWPORT, viewport);
 
-    session->draw(viewport[2], viewport[3]);
+    session->view_draw(viewport[2], viewport[3]);
   }
 
   Py_RETURN_NONE;
@@ -697,40 +706,6 @@ static PyObject *system_info_func(PyObject * /*self*/, PyObject * /*value*/)
   return pyunicode_from_string(system_info.c_str());
 }
 
-#ifdef WITH_OPENCL
-static PyObject *opencl_disable_func(PyObject * /*self*/, PyObject * /*value*/)
-{
-  VLOG(2) << "Disabling OpenCL platform.";
-  DebugFlags().opencl.device_type = DebugFlags::OpenCL::DEVICE_NONE;
-  Py_RETURN_NONE;
-}
-
-static PyObject *opencl_compile_func(PyObject * /*self*/, PyObject *args)
-{
-  PyObject *sequence = PySequence_Fast(args, "Arguments must be a sequence");
-  if (sequence == NULL) {
-    Py_RETURN_FALSE;
-  }
-
-  vector<string> parameters;
-  for (Py_ssize_t i = 0; i < PySequence_Fast_GET_SIZE(sequence); i++) {
-    PyObject *item = PySequence_Fast_GET_ITEM(sequence, i);
-    PyObject *item_as_string = PyObject_Str(item);
-    const char *parameter_string = PyUnicode_AsUTF8(item_as_string);
-    parameters.push_back(parameter_string);
-    Py_DECREF(item_as_string);
-  }
-  Py_DECREF(sequence);
-
-  if (device_opencl_compile_kernel(parameters)) {
-    Py_RETURN_TRUE;
-  }
-  else {
-    Py_RETURN_FALSE;
-  }
-}
-#endif
-
 static bool image_parse_filepaths(PyObject *pyfilepaths, vector<string> &filepaths)
 {
   if (PyUnicode_Check(pyfilepaths)) {
@@ -762,6 +737,10 @@ static bool image_parse_filepaths(PyObject *pyfilepaths, vector<string> &filepat
 
 static PyObject *denoise_func(PyObject * /*self*/, PyObject *args, PyObject *keywords)
 {
+#if 1
+  (void)args;
+  (void)keywords;
+#else
   static const char *keyword_list[] = {
       "preferences", "scene", "view_layer", "input", "output", "tile_size", "samples", NULL};
   PyObject *pypreferences, *pyscene, *pyviewlayer;
@@ -835,7 +814,7 @@ static PyObject *denoise_func(PyObject * /*self*/, PyObject *args, PyObject *key
   }
 
   /* Create denoiser. */
-  Denoiser denoiser(device);
+  DenoiserPipeline denoiser(device);
   denoiser.params = params;
   denoiser.input = input;
   denoiser.output = output;
@@ -852,6 +831,7 @@ static PyObject *denoise_func(PyObject * /*self*/, PyObject *args, PyObject *key
     PyErr_SetString(PyExc_ValueError, denoiser.error.c_str());
     return NULL;
   }
+#endif
 
   Py_RETURN_NONE;
 }
@@ -903,10 +883,7 @@ static PyObject *debug_flags_update_func(PyObject * /*self*/, PyObject *args)
   RNA_id_pointer_create((ID *)PyLong_AsVoidPtr(pyscene), &sceneptr);
   BL::Scene b_scene(sceneptr);
 
-  if (debug_flags_sync_from_scene(b_scene)) {
-    VLOG(2) << "Tagging device list for update.";
-    Device::tag_update();
-  }
+  debug_flags_sync_from_scene(b_scene);
 
   VLOG(2) << "Debug flags set to:\n" << DebugFlags();
 
@@ -917,10 +894,7 @@ static PyObject *debug_flags_update_func(PyObject * /*self*/, PyObject *args)
 
 static PyObject *debug_flags_reset_func(PyObject * /*self*/, PyObject * /*args*/)
 {
-  if (debug_flags_reset()) {
-    VLOG(2) << "Tagging device list for update.";
-    Device::tag_update();
-  }
+  debug_flags_reset();
   if (debug_flags_set) {
     VLOG(2) << "Debug flags reset to:\n" << DebugFlags();
     debug_flags_set = false;
@@ -928,84 +902,6 @@ static PyObject *debug_flags_reset_func(PyObject * /*self*/, PyObject * /*args*/
   Py_RETURN_NONE;
 }
 
-static PyObject *set_resumable_chunk_func(PyObject * /*self*/, PyObject *args)
-{
-  int num_resumable_chunks, current_resumable_chunk;
-  if (!PyArg_ParseTuple(args, "ii", &num_resumable_chunks, &current_resumable_chunk)) {
-    Py_RETURN_NONE;
-  }
-
-  if (num_resumable_chunks <= 0) {
-    fprintf(stderr, "Cycles: Bad value for number of resumable chunks.\n");
-    abort();
-    Py_RETURN_NONE;
-  }
-  if (current_resumable_chunk < 1 || current_resumable_chunk > num_resumable_chunks) {
-    fprintf(stderr, "Cycles: Bad value for current resumable chunk number.\n");
-    abort();
-    Py_RETURN_NONE;
-  }
-
-  VLOG(1) << "Initialized resumable render: "
-          << "num_resumable_chunks=" << num_resumable_chunks << ", "
-          << "current_resumable_chunk=" << current_resumable_chunk;
-  BlenderSession::num_resumable_chunks = num_resumable_chunks;
-  BlenderSession::current_resumable_chunk = current_resumable_chunk;
-
-  printf("Cycles: Will render chunk %d of %d\n", current_resumable_chunk, num_resumable_chunks);
-
-  Py_RETURN_NONE;
-}
-
-static PyObject *set_resumable_chunk_range_func(PyObject * /*self*/, PyObject *args)
-{
-  int num_chunks, start_chunk, end_chunk;
-  if (!PyArg_ParseTuple(args, "iii", &num_chunks, &start_chunk, &end_chunk)) {
-    Py_RETURN_NONE;
-  }
-
-  if (num_chunks <= 0) {
-    fprintf(stderr, "Cycles: Bad value for number of resumable chunks.\n");
-    abort();
-    Py_RETURN_NONE;
-  }
-  if (start_chunk < 1 || start_chunk > num_chunks) {
-    fprintf(stderr, "Cycles: Bad value for start chunk number.\n");
-    abort();
-    Py_RETURN_NONE;
-  }
-  if (end_chunk < 1 || end_chunk > num_chunks) {
-    fprintf(stderr, "Cycles: Bad value for start chunk number.\n");
-    abort();
-    Py_RETURN_NONE;
-  }
-  if (start_chunk > end_chunk) {
-    fprintf(stderr, "Cycles: End chunk should be higher than start one.\n");
-    abort();
-    Py_RETURN_NONE;
-  }
-
-  VLOG(1) << "Initialized resumable render: "
-          << "num_resumable_chunks=" << num_chunks << ", "
-          << "start_resumable_chunk=" << start_chunk << "end_resumable_chunk=" << end_chunk;
-  BlenderSession::num_resumable_chunks = num_chunks;
-  BlenderSession::start_resumable_chunk = start_chunk;
-  BlenderSession::end_resumable_chunk = end_chunk;
-
-  printf("Cycles: Will render chunks %d to %d of %d\n", start_chunk, end_chunk, num_chunks);
-
-  Py_RETURN_NONE;
-}
-
-static PyObject *clear_resumable_chunk_func(PyObject * /*self*/, PyObject * /*value*/)
-{
-  VLOG(1) << "Clear resumable render";
-  BlenderSession::num_resumable_chunks = 0;
-  BlenderSession::current_resumable_chunk = 0;
-
-  Py_RETURN_NONE;
-}
-
 static PyObject *enable_print_stats_func(PyObject * /*self*/, PyObject * /*args*/)
 {
   BlenderSession::print_render_stats = true;
@@ -1015,16 +911,16 @@ static PyObject *enable_print_stats_func(PyObject * /*self*/, PyObject * /*args*
 static PyObject *get_device_types_func(PyObject * /*self*/, PyObject * /*args*/)
 {
   vector<DeviceType> device_types = Device::available_types();
-  bool has_cuda = false, has_optix = false, has_opencl = false;
+  bool has_cuda = false, has_optix = false, has_hip = false;
   foreach (DeviceType device_type, device_types) {
     has_cuda |= (device_type == DEVICE_CUDA);
     has_optix |= (device_type == DEVICE_OPTIX);
-    has_opencl |= (device_type == DEVICE_OPENCL);
+    has_hip |= (device_type == DEVICE_HIP);
   }
   PyObject *list = PyTuple_New(3);
   PyTuple_SET_ITEM(list, 0, PyBool_FromLong(has_cuda));
   PyTuple_SET_ITEM(list, 1, PyBool_FromLong(has_optix));
-  PyTuple_SET_ITEM(list, 2, PyBool_FromLong(has_opencl));
+  PyTuple_SET_ITEM(list, 2, PyBool_FromLong(has_hip));
   return list;
 }
 
@@ -1044,15 +940,15 @@ static PyObject *set_device_override_func(PyObject * /*self*/, PyObject *arg)
   if (override == "CPU") {
     BlenderSession::device_override = DEVICE_MASK_CPU;
   }
-  else if (override == "OPENCL") {
-    BlenderSession::device_override = DEVICE_MASK_OPENCL;
-  }
   else if (override == "CUDA") {
     BlenderSession::device_override = DEVICE_MASK_CUDA;
   }
   else if (override == "OPTIX") {
     BlenderSession::device_override = DEVICE_MASK_OPTIX;
   }
+  else if (override == "HIP") {
+    BlenderSession::device_override = DEVICE_MASK_HIP;
+  }
   else {
     printf("\nError: %s is not a valid Cycles device.\n", override.c_str());
     Py_RETURN_FALSE;
@@ -1072,8 +968,10 @@ static PyMethodDef methods[] = {
     {"create", create_func, METH_VARARGS, ""},
     {"free", free_func, METH_O, ""},
     {"render", render_func, METH_VARARGS, ""},
-    {"bake", bake_func, METH_VARARGS, ""},
+    {"render_frame_finish", render_frame_finish_func, METH_VARARGS, ""},
     {"draw", draw_func, METH_VARARGS, ""},
+    {"bake", bake_func, METH_VARARGS, ""},
+    {"view_draw", view_draw_func, METH_VARARGS, ""},
     {"sync", sync_func, METH_VARARGS, ""},
     {"reset", reset_func, METH_VARARGS, ""},
 #ifdef WITH_OSL
@@ -1082,10 +980,6 @@ static PyMethodDef methods[] = {
 #endif
     {"available_devices", available_devices_func, METH_VARARGS, ""},
     {"system_info", system_info_func, METH_NOARGS, ""},
-#ifdef WITH_OPENCL
-    {"opencl_disable", opencl_disable_func, METH_NOARGS, ""},
-    {"opencl_compile", opencl_compile_func, METH_VARARGS, ""},
-#endif
 
     /* Standalone denoising */
     {"denoise", (PyCFunction)denoise_func, METH_VARARGS | METH_KEYWORDS, ""},
@@ -1098,11 +992,6 @@ static PyMethodDef methods[] = {
     /* Statistics. */
     {"enable_print_stats", enable_print_stats_func, METH_NOARGS, ""},
 
-    /* Resumable render */
-    {"set_resumable_chunk", set_resumable_chunk_func, METH_VARARGS, ""},
-    {"set_resumable_chunk_range", set_resumable_chunk_range_func, METH_VARARGS, ""},
-    {"clear_resumable_chunk", clear_resumable_chunk_func, METH_NOARGS, ""},
-
     /* Compute Device selection */
     {"get_device_types", get_device_types_func, METH_VARARGS, ""},
     {"set_device_override", set_device_override_func, METH_O, ""},
@@ -1153,14 +1042,6 @@ void *CCL_python_module_init()
   PyModule_AddStringConstant(mod, "osl_version_string", "unknown");
 #endif
 
-#ifdef WITH_NETWORK
-  PyModule_AddObject(mod, "with_network", Py_True);
-  Py_INCREF(Py_True);
-#else  /* WITH_NETWORK */
-  PyModule_AddObject(mod, "with_network", Py_False);
-  Py_INCREF(Py_False);
-#endif /* WITH_NETWORK */
-
 #ifdef WITH_EMBREE
   PyModule_AddObject(mod, "with_embree", Py_True);
   Py_INCREF(Py_True);

intern/cycles/blender/blender_session.h

@@ -29,12 +29,11 @@
 
 CCL_NAMESPACE_BEGIN
 
+class BlenderGPUDisplay;
 class BlenderSync;
 class ImageMetaData;
 class Scene;
 class Session;
-class RenderBuffers;
-class RenderTile;
 
 class BlenderSession {
  public:
@@ -62,6 +61,8 @@ class BlenderSession {
   /* offline render */
   void render(BL::Depsgraph &b_depsgraph);
 
+  void render_frame_finish();
+
   void bake(BL::Depsgraph &b_depsgrah,
             BL::Object &b_object,
             const string &pass_type,
@@ -69,24 +70,29 @@ class BlenderSession {
             const int bake_width,
             const int bake_height);
 
-  void write_render_result(BL::RenderLayer &b_rlay, RenderTile &rtile);
-  void write_render_tile(RenderTile &rtile);
-  void read_render_tile(RenderTile &rtile);
+  void write_render_result(BL::RenderLayer &b_rlay);
+  void write_render_tile();
+
+  void update_render_tile();
+
+  void full_buffer_written(string_view filename);
 
   /* update functions are used to update display buffer only after sample was rendered
    * only needed for better visual feedback */
-  void update_render_result(BL::RenderLayer &b_rlay, RenderTile &rtile);
-  void update_render_tile(RenderTile &rtile, bool highlight);
+  void update_render_result(BL::RenderLayer &b_rlay);
+
+  /* read functions for baking input */
+  void read_render_tile();
 
   /* interactive updates */
   void synchronize(BL::Depsgraph &b_depsgraph);
 
   /* drawing */
-  bool draw(int w, int h);
+  void draw(BL::SpaceImageEditor &space_image);
+  void view_draw(int w, int h);
   void tag_redraw();
   void tag_update();
   void get_status(string &status, string &substatus);
-  void get_kernel_status(string &kernel_status);
   void get_progress(float &progress, double &total_time, double &render_time);
   void test_cancel();
   void update_status_progress();
@@ -104,8 +110,7 @@ class BlenderSession {
   BL::RenderSettings b_render;
   BL::Depsgraph b_depsgraph;
   /* NOTE: Blender's scene might become invalid after call
-   * free_blender_memory_if_possible().
-   */
+   * #free_blender_memory_if_possible(). */
   BL::Scene b_scene;
   BL::SpaceView3D b_v3d;
   BL::RegionView3D b_rv3d;
@@ -123,6 +128,8 @@ class BlenderSession {
 
   void *python_thread_state;
 
+  bool use_developer_ui;
+
   /* Global state which is common for all render sessions created from Blender.
    * Usually denotes command line arguments.
    */
@@ -134,41 +141,33 @@ class BlenderSession {
    */
   static bool headless;
 
-  /* ** Resumable render ** */
-
-  /* Overall number of chunks in which the sample range is to be divided. */
-  static int num_resumable_chunks;
-
-  /* Current resumable chunk index to render. */
-  static int current_resumable_chunk;
-
-  /* Alternative to single-chunk rendering to render a range of chunks. */
-  static int start_resumable_chunk;
-  static int end_resumable_chunk;
-
   static bool print_render_stats;
 
  protected:
   void stamp_view_layer_metadata(Scene *scene, const string &view_layer_name);
 
-  void do_write_update_render_result(BL::RenderLayer &b_rlay,
-                                     RenderTile &rtile,
-                                     bool do_update_only);
-  void do_write_update_render_tile(RenderTile &rtile,
-                                   bool do_update_only,
-                                   bool do_read_only,
-                                   bool highlight);
+  /* Check whether session error happened.
+   * If so, it is reported to the render engine and true is returned.
+   * Otherwise false is returned. */
+  bool check_and_report_session_error();
 
   void builtin_images_load();
 
-  /* Update tile manager to reflect resumable render settings. */
-  void update_resumable_tile_manager(int num_samples);
-
   /* Is used after each render layer synchronization is done with the goal
    * of freeing render engine data which is held from Blender side (for
    * example, dependency graph).
    */
   void free_blender_memory_if_possible();
+
+  struct {
+    thread_mutex mutex;
+    int last_pass_index = -1;
+  } draw_state_;
+
+  /* NOTE: The BlenderSession references the GPU display. */
+  BlenderGPUDisplay *gpu_display_ = nullptr;
+
+  vector<string> full_buffer_files_;
 };
 
 CCL_NAMESPACE_END

intern/cycles/blender/blender_shader.cpp

@@ -17,6 +17,7 @@
 #include "render/background.h"
 #include "render/colorspace.h"
 #include "render/graph.h"
+#include "render/integrator.h"
 #include "render/light.h"
 #include "render/nodes.h"
 #include "render/osl.h"
@@ -475,17 +476,11 @@ static ShaderNode *add_node(Scene *scene,
     SubsurfaceScatteringNode *subsurface = graph->create_node<SubsurfaceScatteringNode>();
 
     switch (b_subsurface_node.falloff()) {
-      case BL::ShaderNodeSubsurfaceScattering::falloff_CUBIC:
-        subsurface->set_falloff(CLOSURE_BSSRDF_CUBIC_ID);
-        break;
-      case BL::ShaderNodeSubsurfaceScattering::falloff_GAUSSIAN:
-        subsurface->set_falloff(CLOSURE_BSSRDF_GAUSSIAN_ID);
-        break;
-      case BL::ShaderNodeSubsurfaceScattering::falloff_BURLEY:
-        subsurface->set_falloff(CLOSURE_BSSRDF_BURLEY_ID);
+      case BL::ShaderNodeSubsurfaceScattering::falloff_RANDOM_WALK_FIXED_RADIUS:
+        subsurface->set_method(CLOSURE_BSSRDF_RANDOM_WALK_FIXED_RADIUS_ID);
         break;
       case BL::ShaderNodeSubsurfaceScattering::falloff_RANDOM_WALK:
-        subsurface->set_falloff(CLOSURE_BSSRDF_RANDOM_WALK_ID);
+        subsurface->set_method(CLOSURE_BSSRDF_RANDOM_WALK_ID);
         break;
     }
 
@@ -597,11 +592,11 @@ static ShaderNode *add_node(Scene *scene,
         break;
     }
     switch (b_principled_node.subsurface_method()) {
-      case BL::ShaderNodeBsdfPrincipled::subsurface_method_BURLEY:
-        principled->set_subsurface_method(CLOSURE_BSSRDF_PRINCIPLED_ID);
+      case BL::ShaderNodeBsdfPrincipled::subsurface_method_RANDOM_WALK_FIXED_RADIUS:
+        principled->set_subsurface_method(CLOSURE_BSSRDF_RANDOM_WALK_FIXED_RADIUS_ID);
         break;
       case BL::ShaderNodeBsdfPrincipled::subsurface_method_RANDOM_WALK:
-        principled->set_subsurface_method(CLOSURE_BSSRDF_PRINCIPLED_RANDOM_WALK_ID);
+        principled->set_subsurface_method(CLOSURE_BSSRDF_RANDOM_WALK_ID);
         break;
     }
     node = principled;
@@ -1360,10 +1355,11 @@ void BlenderSync::sync_materials(BL::Depsgraph &b_depsgraph, bool update_all)
 void BlenderSync::sync_world(BL::Depsgraph &b_depsgraph, BL::SpaceView3D &b_v3d, bool update_all)
 {
   Background *background = scene->background;
+  Integrator *integrator = scene->integrator;
 
   BL::World b_world = b_scene.world();
 
-  BlenderViewportParameters new_viewport_parameters(b_v3d);
+  BlenderViewportParameters new_viewport_parameters(b_v3d, use_developer_ui);
 
   if (world_recalc || update_all || b_world.ptr.data != world_map ||
       viewport_parameters.shader_modified(new_viewport_parameters)) {
@@ -1455,9 +1451,8 @@ void BlenderSync::sync_world(BL::Depsgraph &b_depsgraph, BL::SpaceView3D &b_v3d,
       /* AO */
       BL::WorldLighting b_light = b_world.light_settings();
 
-      background->set_use_ao(b_light.use_ambient_occlusion());
-      background->set_ao_factor(b_light.ao_factor());
-      background->set_ao_distance(b_light.distance());
+      integrator->set_ao_factor(b_light.ao_factor());
+      integrator->set_ao_distance(b_light.distance());
 
       /* visibility */
       PointerRNA cvisibility = RNA_pointer_get(&b_world.ptr, "cycles_visibility");
@@ -1472,9 +1467,8 @@ void BlenderSync::sync_world(BL::Depsgraph &b_depsgraph, BL::SpaceView3D &b_v3d,
       background->set_visibility(visibility);
     }
     else {
-      background->set_use_ao(false);
-      background->set_ao_factor(0.0f);
-      background->set_ao_distance(FLT_MAX);
+      integrator->set_ao_factor(1.0f);
+      integrator->set_ao_distance(10.0f);
     }
 
     shader->set_graph(graph);
@@ -1496,7 +1490,6 @@ void BlenderSync::sync_world(BL::Depsgraph &b_depsgraph, BL::SpaceView3D &b_v3d,
 
   background->set_use_shader(view_layer.use_background_shader ||
                              viewport_parameters.use_custom_shader());
-  background->set_use_ao(background->get_use_ao() && view_layer.use_background_ao);
 
   background->tag_update(scene);
 }

intern/cycles/blender/blender_sync.cpp

@@ -53,6 +53,7 @@ BlenderSync::BlenderSync(BL::RenderEngine &b_engine,
                          BL::Scene &b_scene,
                          Scene *scene,
                          bool preview,
+                         bool use_developer_ui,
                          Progress &progress)
     : b_engine(b_engine),
       b_data(b_data),
@@ -68,6 +69,7 @@ BlenderSync::BlenderSync(BL::RenderEngine &b_engine,
       scene(scene),
       preview(preview),
       experimental(false),
+      use_developer_ui(use_developer_ui),
       dicing_rate(1.0f),
       max_subdivisions(12),
       progress(progress),
@@ -224,7 +226,7 @@ void BlenderSync::sync_recalc(BL::Depsgraph &b_depsgraph, BL::SpaceView3D &b_v3d
   }
 
   if (b_v3d) {
-    BlenderViewportParameters new_viewport_parameters(b_v3d);
+    BlenderViewportParameters new_viewport_parameters(b_v3d, use_developer_ui);
 
     if (viewport_parameters.shader_modified(new_viewport_parameters)) {
       world_recalc = true;
@@ -251,9 +253,13 @@ void BlenderSync::sync_data(BL::RenderSettings &b_render,
 
   BL::ViewLayer b_view_layer = b_depsgraph.view_layer_eval();
 
+  /* TODO(sergey): This feels weak to pass view layer to the integrator, and even weaker to have an
+   * implicit check on whether it is a background render or not. What is the nicer thing here? */
+  const bool background = !b_v3d;
+
   sync_view_layer(b_view_layer);
-  sync_integrator();
-  sync_film(b_v3d);
+  sync_integrator(b_view_layer, background);
+  sync_film(b_view_layer, b_v3d);
   sync_shaders(b_depsgraph, b_v3d);
   sync_images();
 
@@ -280,7 +286,7 @@ void BlenderSync::sync_data(BL::RenderSettings &b_render,
 
 /* Integrator */
 
-void BlenderSync::sync_integrator()
+void BlenderSync::sync_integrator(BL::ViewLayer &b_view_layer, bool background)
 {
   PointerRNA cscene = RNA_pointer_get(&b_scene.ptr, "cycles");
 
@@ -328,59 +334,24 @@ void BlenderSync::sync_integrator()
     integrator->set_motion_blur(view_layer.use_motion_blur);
   }
 
-  integrator->set_method((Integrator::Method)get_enum(
-      cscene, "progressive", Integrator::NUM_METHODS, Integrator::PATH));
-
-  integrator->set_sample_all_lights_direct(get_boolean(cscene, "sample_all_lights_direct"));
-  integrator->set_sample_all_lights_indirect(get_boolean(cscene, "sample_all_lights_indirect"));
   integrator->set_light_sampling_threshold(get_float(cscene, "light_sampling_threshold"));
 
   SamplingPattern sampling_pattern = (SamplingPattern)get_enum(
       cscene, "sampling_pattern", SAMPLING_NUM_PATTERNS, SAMPLING_PATTERN_SOBOL);
-
-  int adaptive_min_samples = INT_MAX;
-
-  if (RNA_boolean_get(&cscene, "use_adaptive_sampling")) {
-    sampling_pattern = SAMPLING_PATTERN_PMJ;
-    adaptive_min_samples = get_int(cscene, "adaptive_min_samples");
-    integrator->set_adaptive_threshold(get_float(cscene, "adaptive_threshold"));
-  }
-  else {
-    integrator->set_adaptive_threshold(0.0f);
-  }
-
   integrator->set_sampling_pattern(sampling_pattern);
 
-  int diffuse_samples = get_int(cscene, "diffuse_samples");
-  int glossy_samples = get_int(cscene, "glossy_samples");
-  int transmission_samples = get_int(cscene, "transmission_samples");
-  int ao_samples = get_int(cscene, "ao_samples");
-  int mesh_light_samples = get_int(cscene, "mesh_light_samples");
-  int subsurface_samples = get_int(cscene, "subsurface_samples");
-  int volume_samples = get_int(cscene, "volume_samples");
-
-  if (get_boolean(cscene, "use_square_samples")) {
-    integrator->set_diffuse_samples(diffuse_samples * diffuse_samples);
-    integrator->set_glossy_samples(glossy_samples * glossy_samples);
-    integrator->set_transmission_samples(transmission_samples * transmission_samples);
-    integrator->set_ao_samples(ao_samples * ao_samples);
-    integrator->set_mesh_light_samples(mesh_light_samples * mesh_light_samples);
-    integrator->set_subsurface_samples(subsurface_samples * subsurface_samples);
-    integrator->set_volume_samples(volume_samples * volume_samples);
-    adaptive_min_samples = min(adaptive_min_samples * adaptive_min_samples, INT_MAX);
+  if (preview) {
+    integrator->set_use_adaptive_sampling(
+        RNA_boolean_get(&cscene, "use_preview_adaptive_sampling"));
+    integrator->set_adaptive_threshold(get_float(cscene, "preview_adaptive_threshold"));
+    integrator->set_adaptive_min_samples(get_int(cscene, "preview_adaptive_min_samples"));
   }
   else {
-    integrator->set_diffuse_samples(diffuse_samples);
-    integrator->set_glossy_samples(glossy_samples);
-    integrator->set_transmission_samples(transmission_samples);
-    integrator->set_ao_samples(ao_samples);
-    integrator->set_mesh_light_samples(mesh_light_samples);
-    integrator->set_subsurface_samples(subsurface_samples);
-    integrator->set_volume_samples(volume_samples);
+    integrator->set_use_adaptive_sampling(RNA_boolean_get(&cscene, "use_adaptive_sampling"));
+    integrator->set_adaptive_threshold(get_float(cscene, "adaptive_threshold"));
+    integrator->set_adaptive_min_samples(get_int(cscene, "adaptive_min_samples"));
   }
 
-  integrator->set_adaptive_min_samples(adaptive_min_samples);
-
   if (get_boolean(cscene, "use_fast_gi")) {
     if (preview) {
       integrator->set_ao_bounces(get_int(cscene, "ao_bounces"));
@@ -393,20 +364,38 @@ void BlenderSync::sync_integrator()
     integrator->set_ao_bounces(0);
   }
 
-  /* UPDATE_NONE as we don't want to tag the integrator as modified, just tag dependent things */
+  const DenoiseParams denoise_params = get_denoise_params(b_scene, b_view_layer, background);
+  integrator->set_use_denoise(denoise_params.use);
+
+  /* Only update denoiser parameters if the denoiser is actually used. This allows to tweak
+   * denoiser parameters before enabling it without render resetting on every change. The downside
+   * is that the interface and the integrator are technically out of sync. */
+  if (denoise_params.use) {
+    integrator->set_denoiser_type(denoise_params.type);
+    integrator->set_denoise_start_sample(denoise_params.start_sample);
+    integrator->set_use_denoise_pass_albedo(denoise_params.use_pass_albedo);
+    integrator->set_use_denoise_pass_normal(denoise_params.use_pass_normal);
+    integrator->set_denoiser_prefilter(denoise_params.prefilter);
+  }
+
+  /* UPDATE_NONE as we don't want to tag the integrator as modified (this was done by the
+   * set calls above), but we need to make sure that the dependent things are tagged. */
   integrator->tag_update(scene, Integrator::UPDATE_NONE);
 }
 
 /* Film */
 
-void BlenderSync::sync_film(BL::SpaceView3D &b_v3d)
+void BlenderSync::sync_film(BL::ViewLayer &b_view_layer, BL::SpaceView3D &b_v3d)
 {
   PointerRNA cscene = RNA_pointer_get(&b_scene.ptr, "cycles");
+  PointerRNA crl = RNA_pointer_get(&b_view_layer.ptr, "cycles");
 
   Film *film = scene->film;
 
   if (b_v3d) {
-    film->set_display_pass(update_viewport_display_passes(b_v3d, scene->passes));
+    const BlenderViewportParameters new_viewport_parameters(b_v3d, use_developer_ui);
+    film->set_display_pass(new_viewport_parameters.display_pass);
+    film->set_show_active_pixels(new_viewport_parameters.show_active_pixels);
   }
 
   film->set_exposure(get_float(cscene, "film_exposure"));
@@ -434,6 +423,15 @@ void BlenderSync::sync_film(BL::SpaceView3D &b_v3d)
         break;
     }
   }
+
+  /* Blender viewport does not support proper shadow catcher compositing, so force an approximate
+   * mode to improve visual feedback. */
+  if (b_v3d) {
+    film->set_use_approximate_shadow_catcher(true);
+  }
+  else {
+    film->set_use_approximate_shadow_catcher(!get_boolean(crl, "use_pass_shadow_catcher"));
+  }
 }
 
 /* Render Layer */
@@ -444,7 +442,6 @@ void BlenderSync::sync_view_layer(BL::ViewLayer &b_view_layer)
 
   /* Filter. */
   view_layer.use_background_shader = b_view_layer.use_sky();
-  view_layer.use_background_ao = b_view_layer.use_ao();
   /* Always enable surfaces for baking, otherwise there is nothing to bake to. */
   view_layer.use_surfaces = b_view_layer.use_solid() || scene->bake_manager->get_baking();
   view_layer.use_hair = b_view_layer.use_strand();
@@ -464,10 +461,7 @@ void BlenderSync::sync_view_layer(BL::ViewLayer &b_view_layer)
 
   if (use_layer_samples != 2) {
     int samples = b_view_layer.samples();
-    if (get_boolean(cscene, "use_square_samples"))
-      view_layer.samples = samples * samples;
-    else
-      view_layer.samples = samples;
+    view_layer.samples = samples;
   }
 }
 
@@ -499,7 +493,8 @@ void BlenderSync::sync_images()
 }
 
 /* Passes */
-PassType BlenderSync::get_pass_type(BL::RenderPass &b_pass)
+
+static PassType get_blender_pass_type(BL::RenderPass &b_pass)
 {
   string name = b_pass.name();
 #define MAP_PASS(passname, passtype) \
@@ -507,10 +502,15 @@ PassType BlenderSync::get_pass_type(BL::RenderPass &b_pass)
     return passtype; \
   } \
   ((void)0)
+
   /* NOTE: Keep in sync with defined names from DNA_scene_types.h */
+
   MAP_PASS("Combined", PASS_COMBINED);
+  MAP_PASS("Noisy Image", PASS_COMBINED);
+
   MAP_PASS("Depth", PASS_DEPTH);
   MAP_PASS("Mist", PASS_MIST);
+  MAP_PASS("Position", PASS_POSITION);
   MAP_PASS("Normal", PASS_NORMAL);
   MAP_PASS("IndexOB", PASS_OBJECT_ID);
   MAP_PASS("UV", PASS_UV);
@@ -539,118 +539,92 @@ PassType BlenderSync::get_pass_type(BL::RenderPass &b_pass)
   MAP_PASS("BakePrimitive", PASS_BAKE_PRIMITIVE);
   MAP_PASS("BakeDifferential", PASS_BAKE_DIFFERENTIAL);
 
+  MAP_PASS("Denoising Normal", PASS_DENOISING_NORMAL);
+  MAP_PASS("Denoising Albedo", PASS_DENOISING_ALBEDO);
+
+  MAP_PASS("Shadow Catcher", PASS_SHADOW_CATCHER);
+  MAP_PASS("Noisy Shadow Catcher", PASS_SHADOW_CATCHER);
+
   MAP_PASS("Debug Render Time", PASS_RENDER_TIME);
+
   MAP_PASS("AdaptiveAuxBuffer", PASS_ADAPTIVE_AUX_BUFFER);
   MAP_PASS("Debug Sample Count", PASS_SAMPLE_COUNT);
+
   if (string_startswith(name, cryptomatte_prefix)) {
     return PASS_CRYPTOMATTE;
   }
+
 #undef MAP_PASS
 
   return PASS_NONE;
 }
 
-int BlenderSync::get_denoising_pass(BL::RenderPass &b_pass)
+static Pass *pass_add(Scene *scene,
+                      PassType type,
+                      const char *name,
+                      PassMode mode = PassMode::DENOISED)
 {
-  string name = b_pass.name();
+  Pass *pass = scene->create_node<Pass>();
 
-  if (name == "Noisy Image")
-    return DENOISING_PASS_PREFILTERED_COLOR;
+  pass->set_type(type);
+  pass->set_name(ustring(name));
+  pass->set_mode(mode);
 
-  if (name.substr(0, 10) != "Denoising ") {
-    return -1;
-  }
-  name = name.substr(10);
-
-#define MAP_PASS(passname, offset) \
-  if (name == passname) { \
-    return offset; \
-  } \
-  ((void)0)
-  MAP_PASS("Normal", DENOISING_PASS_PREFILTERED_NORMAL);
-  MAP_PASS("Albedo", DENOISING_PASS_PREFILTERED_ALBEDO);
-  MAP_PASS("Depth", DENOISING_PASS_PREFILTERED_DEPTH);
-  MAP_PASS("Shadowing", DENOISING_PASS_PREFILTERED_SHADOWING);
-  MAP_PASS("Variance", DENOISING_PASS_PREFILTERED_VARIANCE);
-  MAP_PASS("Intensity", DENOISING_PASS_PREFILTERED_INTENSITY);
-  MAP_PASS("Clean", DENOISING_PASS_CLEAN);
-#undef MAP_PASS
-
-  return -1;
+  return pass;
 }
 
-vector<Pass> BlenderSync::sync_render_passes(BL::Scene &b_scene,
-                                             BL::RenderLayer &b_rlay,
-                                             BL::ViewLayer &b_view_layer,
-                                             bool adaptive_sampling,
-                                             const DenoiseParams &denoising)
+void BlenderSync::sync_render_passes(BL::RenderLayer &b_rlay, BL::ViewLayer &b_view_layer)
 {
-  vector<Pass> passes;
+  PointerRNA cscene = RNA_pointer_get(&b_scene.ptr, "cycles");
 
-  /* loop over passes */
+  /* Delete all existing passes. */
+  set<Pass *> clear_passes(scene->passes.begin(), scene->passes.end());
+  scene->delete_nodes(clear_passes);
+
+  /* Always add combined pass. */
+  pass_add(scene, PASS_COMBINED, "Combined");
+
+  /* Blender built-in data and light passes. */
   for (BL::RenderPass &b_pass : b_rlay.passes) {
-    PassType pass_type = get_pass_type(b_pass);
+    const PassType pass_type = get_blender_pass_type(b_pass);
+
+    if (pass_type == PASS_NONE) {
+      LOG(ERROR) << "Unknown pass " << b_pass.name();
+      continue;
+    }
 
     if (pass_type == PASS_MOTION &&
         (b_view_layer.use_motion_blur() && b_scene.render().use_motion_blur())) {
       continue;
     }
-    if (pass_type != PASS_NONE)
-      Pass::add(pass_type, passes, b_pass.name().c_str());
+
+    pass_add(scene, pass_type, b_pass.name().c_str());
   }
 
   PointerRNA crl = RNA_pointer_get(&b_view_layer.ptr, "cycles");
 
-  int denoising_flags = 0;
-  if (denoising.use || denoising.store_passes) {
-    if (denoising.type == DENOISER_NLM) {
-#define MAP_OPTION(name, flag) \
-  if (!get_boolean(crl, name)) { \
-    denoising_flags |= flag; \
-  } \
-  ((void)0)
-      MAP_OPTION("denoising_diffuse_direct", DENOISING_CLEAN_DIFFUSE_DIR);
-      MAP_OPTION("denoising_diffuse_indirect", DENOISING_CLEAN_DIFFUSE_IND);
-      MAP_OPTION("denoising_glossy_direct", DENOISING_CLEAN_GLOSSY_DIR);
-      MAP_OPTION("denoising_glossy_indirect", DENOISING_CLEAN_GLOSSY_IND);
-      MAP_OPTION("denoising_transmission_direct", DENOISING_CLEAN_TRANSMISSION_DIR);
-      MAP_OPTION("denoising_transmission_indirect", DENOISING_CLEAN_TRANSMISSION_IND);
-#undef MAP_OPTION
-    }
-    b_engine.add_pass("Noisy Image", 4, "RGBA", b_view_layer.name().c_str());
-  }
-  scene->film->set_denoising_flags(denoising_flags);
-
-  if (denoising.store_passes) {
-    b_engine.add_pass("Denoising Normal", 3, "XYZ", b_view_layer.name().c_str());
-    b_engine.add_pass("Denoising Albedo", 3, "RGB", b_view_layer.name().c_str());
-    b_engine.add_pass("Denoising Depth", 1, "Z", b_view_layer.name().c_str());
-    if (denoising.type == DENOISER_NLM) {
-      b_engine.add_pass("Denoising Shadowing", 1, "X", b_view_layer.name().c_str());
-      b_engine.add_pass("Denoising Variance", 3, "RGB", b_view_layer.name().c_str());
-      b_engine.add_pass("Denoising Intensity", 1, "X", b_view_layer.name().c_str());
-    }
-
-    if (scene->film->get_denoising_flags() & DENOISING_CLEAN_ALL_PASSES) {
-      b_engine.add_pass("Denoising Clean", 3, "RGB", b_view_layer.name().c_str());
-    }
-  }
-
+  /* Debug passes. */
   if (get_boolean(crl, "pass_debug_render_time")) {
     b_engine.add_pass("Debug Render Time", 1, "X", b_view_layer.name().c_str());
-    Pass::add(PASS_RENDER_TIME, passes, "Debug Render Time");
+    pass_add(scene, PASS_RENDER_TIME, "Debug Render Time");
   }
   if (get_boolean(crl, "pass_debug_sample_count")) {
     b_engine.add_pass("Debug Sample Count", 1, "X", b_view_layer.name().c_str());
-    Pass::add(PASS_SAMPLE_COUNT, passes, "Debug Sample Count");
+    pass_add(scene, PASS_SAMPLE_COUNT, "Debug Sample Count");
   }
+
+  /* Cycles specific passes. */
   if (get_boolean(crl, "use_pass_volume_direct")) {
     b_engine.add_pass("VolumeDir", 3, "RGB", b_view_layer.name().c_str());
-    Pass::add(PASS_VOLUME_DIRECT, passes, "VolumeDir");
+    pass_add(scene, PASS_VOLUME_DIRECT, "VolumeDir");
   }
   if (get_boolean(crl, "use_pass_volume_indirect")) {
     b_engine.add_pass("VolumeInd", 3, "RGB", b_view_layer.name().c_str());
-    Pass::add(PASS_VOLUME_INDIRECT, passes, "VolumeInd");
+    pass_add(scene, PASS_VOLUME_INDIRECT, "VolumeInd");
+  }
+  if (get_boolean(crl, "use_pass_shadow_catcher")) {
+    b_engine.add_pass("Shadow Catcher", 3, "RGB", b_view_layer.name().c_str());
+    pass_add(scene, PASS_SHADOW_CATCHER, "Shadow Catcher");
   }
 
   /* Cryptomatte stores two ID/weight pairs per RGBA layer.
@@ -662,7 +636,7 @@ vector<Pass> BlenderSync::sync_render_passes(BL::Scene &b_scene,
     for (int i = 0; i < crypto_depth; i++) {
       string passname = cryptomatte_prefix + string_printf("Object%02d", i);
       b_engine.add_pass(passname.c_str(), 4, "RGBA", b_view_layer.name().c_str());
-      Pass::add(PASS_CRYPTOMATTE, passes, passname.c_str());
+      pass_add(scene, PASS_CRYPTOMATTE, passname.c_str());
     }
     cryptomatte_passes = (CryptomatteType)(cryptomatte_passes | CRYPT_OBJECT);
   }
@@ -670,7 +644,7 @@ vector<Pass> BlenderSync::sync_render_passes(BL::Scene &b_scene,
     for (int i = 0; i < crypto_depth; i++) {
       string passname = cryptomatte_prefix + string_printf("Material%02d", i);
       b_engine.add_pass(passname.c_str(), 4, "RGBA", b_view_layer.name().c_str());
-      Pass::add(PASS_CRYPTOMATTE, passes, passname.c_str());
+      pass_add(scene, PASS_CRYPTOMATTE, passname.c_str());
     }
     cryptomatte_passes = (CryptomatteType)(cryptomatte_passes | CRYPT_MATERIAL);
   }
@@ -678,22 +652,33 @@ vector<Pass> BlenderSync::sync_render_passes(BL::Scene &b_scene,
     for (int i = 0; i < crypto_depth; i++) {
       string passname = cryptomatte_prefix + string_printf("Asset%02d", i);
       b_engine.add_pass(passname.c_str(), 4, "RGBA", b_view_layer.name().c_str());
-      Pass::add(PASS_CRYPTOMATTE, passes, passname.c_str());
+      pass_add(scene, PASS_CRYPTOMATTE, passname.c_str());
     }
     cryptomatte_passes = (CryptomatteType)(cryptomatte_passes | CRYPT_ASSET);
   }
-  if (b_view_layer.use_pass_cryptomatte_accurate() && cryptomatte_passes != CRYPT_NONE) {
-    cryptomatte_passes = (CryptomatteType)(cryptomatte_passes | CRYPT_ACCURATE);
-  }
   scene->film->set_cryptomatte_passes(cryptomatte_passes);
 
-  if (adaptive_sampling) {
-    Pass::add(PASS_ADAPTIVE_AUX_BUFFER, passes);
-    if (!get_boolean(crl, "pass_debug_sample_count")) {
-      Pass::add(PASS_SAMPLE_COUNT, passes);
+  /* Denoising passes. */
+  const bool use_denoising = get_boolean(cscene, "use_denoising") &&
+                             get_boolean(crl, "use_denoising");
+  const bool store_denoising_passes = get_boolean(crl, "denoising_store_passes");
+  if (use_denoising) {
+    b_engine.add_pass("Noisy Image", 4, "RGBA", b_view_layer.name().c_str());
+    pass_add(scene, PASS_COMBINED, "Noisy Image", PassMode::NOISY);
+    if (get_boolean(crl, "use_pass_shadow_catcher")) {
+      b_engine.add_pass("Noisy Shadow Catcher", 3, "RGB", b_view_layer.name().c_str());
+      pass_add(scene, PASS_SHADOW_CATCHER, "Noisy Shadow Catcher", PassMode::NOISY);
     }
   }
+  if (store_denoising_passes) {
+    b_engine.add_pass("Denoising Normal", 3, "XYZ", b_view_layer.name().c_str());
+    pass_add(scene, PASS_DENOISING_NORMAL, "Denoising Normal", PassMode::NOISY);
 
+    b_engine.add_pass("Denoising Albedo", 3, "RGB", b_view_layer.name().c_str());
+    pass_add(scene, PASS_DENOISING_ALBEDO, "Denoising Albedo", PassMode::NOISY);
+  }
+
+  /* Custom AOV passes. */
   BL::ViewLayer::aovs_iterator b_aov_iter;
   for (b_view_layer.aovs.begin(b_aov_iter); b_aov_iter != b_view_layer.aovs.end(); ++b_aov_iter) {
     BL::AOV b_aov(*b_aov_iter);
@@ -706,28 +691,15 @@ vector<Pass> BlenderSync::sync_render_passes(BL::Scene &b_scene,
 
     if (is_color) {
       b_engine.add_pass(name.c_str(), 4, "RGBA", b_view_layer.name().c_str());
-      Pass::add(PASS_AOV_COLOR, passes, name.c_str());
+      pass_add(scene, PASS_AOV_COLOR, name.c_str());
     }
     else {
       b_engine.add_pass(name.c_str(), 1, "X", b_view_layer.name().c_str());
-      Pass::add(PASS_AOV_VALUE, passes, name.c_str());
+      pass_add(scene, PASS_AOV_VALUE, name.c_str());
     }
   }
 
-  scene->film->set_denoising_data_pass(denoising.use || denoising.store_passes);
-  scene->film->set_denoising_clean_pass(scene->film->get_denoising_flags() &
-                                        DENOISING_CLEAN_ALL_PASSES);
-  scene->film->set_denoising_prefiltered_pass(denoising.store_passes &&
-                                              denoising.type == DENOISER_NLM);
   scene->film->set_pass_alpha_threshold(b_view_layer.pass_alpha_threshold());
-
-  if (!Pass::equals(passes, scene->passes)) {
-    scene->film->tag_passes_update(scene, passes);
-    scene->film->tag_modified();
-    scene->integrator->tag_update(scene, Integrator::UPDATE_ALL);
-  }
-
-  return passes;
 }
 
 void BlenderSync::free_data_after_sync(BL::Depsgraph &b_depsgraph)
@@ -773,9 +745,9 @@ SceneParams BlenderSync::get_scene_params(BL::Scene &b_scene, bool background)
     params.shadingsystem = SHADINGSYSTEM_OSL;
 
   if (background || DebugFlags().viewport_static_bvh)
-    params.bvh_type = SceneParams::BVH_STATIC;
+    params.bvh_type = BVH_TYPE_STATIC;
   else
-    params.bvh_type = SceneParams::BVH_DYNAMIC;
+    params.bvh_type = BVH_TYPE_DYNAMIC;
 
   params.use_bvh_spatial_split = RNA_boolean_get(&cscene, "debug_use_spatial_splits");
   params.use_bvh_unaligned_nodes = RNA_boolean_get(&cscene, "debug_use_hair_bvh");
@@ -818,8 +790,7 @@ bool BlenderSync::get_session_pause(BL::Scene &b_scene, bool background)
 SessionParams BlenderSync::get_session_params(BL::RenderEngine &b_engine,
                                               BL::Preferences &b_preferences,
                                               BL::Scene &b_scene,
-                                              bool background,
-                                              BL::ViewLayer b_view_layer)
+                                              bool background)
 {
   SessionParams params;
   PointerRNA cscene = RNA_pointer_get(&b_scene.ptr, "cycles");
@@ -827,7 +798,8 @@ SessionParams BlenderSync::get_session_params(BL::RenderEngine &b_engine,
   /* feature set */
   params.experimental = (get_enum(cscene, "feature_set") != 0);
 
-  /* Background */
+  /* Headless and background rendering. */
+  params.headless = BlenderSession::headless;
   params.background = background;
 
   /* Device */
@@ -836,111 +808,26 @@ SessionParams BlenderSync::get_session_params(BL::RenderEngine &b_engine,
 
   /* samples */
   int samples = get_int(cscene, "samples");
-  int aa_samples = get_int(cscene, "aa_samples");
   int preview_samples = get_int(cscene, "preview_samples");
-  int preview_aa_samples = get_int(cscene, "preview_aa_samples");
 
-  if (get_boolean(cscene, "use_square_samples")) {
-    aa_samples = aa_samples * aa_samples;
-    preview_aa_samples = preview_aa_samples * preview_aa_samples;
-
-    samples = samples * samples;
-    preview_samples = preview_samples * preview_samples;
-  }
-
-  if (get_enum(cscene, "progressive") == 0 && params.device.has_branched_path) {
-    if (background) {
-      params.samples = aa_samples;
-    }
-    else {
-      params.samples = preview_aa_samples;
-      if (params.samples == 0)
-        params.samples = INT_MAX;
-    }
+  if (background) {
+    params.samples = samples;
   }
   else {
-    if (background) {
-      params.samples = samples;
-    }
-    else {
-      params.samples = preview_samples;
-      if (params.samples == 0)
-        params.samples = INT_MAX;
-    }
+    params.samples = preview_samples;
+    if (params.samples == 0)
+      params.samples = INT_MAX;
   }
 
   /* 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) {
-    /* currently GPU could be much slower than CPU when using tiles,
-     * still need to be investigated, but meanwhile make it possible
-     * to work in viewport smoothly
-     */
-    int debug_tile_size = get_int(cscene, "debug_tile_size");
-
-    params.tile_size = make_int2(debug_tile_size, debug_tile_size);
-  }
-  else {
-    int tile_x = b_engine.tile_x();
-    int tile_y = b_engine.tile_y();
-
-    params.tile_size = make_int2(tile_x, tile_y);
-  }
-
-  if ((BlenderSession::headless == false) && background) {
-    params.tile_order = (TileOrder)get_enum(cscene, "tile_order");
-  }
-  else {
-    params.tile_order = TILE_BOTTOM_TO_TOP;
-  }
-
-  /* Denoising */
-  params.denoising = get_denoise_params(b_scene, b_view_layer, background);
-
-  if (params.denoising.use) {
-    /* Add additional denoising devices if we are rendering and denoising
-     * with different devices. */
-    params.device.add_denoising_devices(params.denoising.type);
-
-    /* Check if denoiser is supported by device. */
-    if (!(params.device.denoisers & params.denoising.type)) {
-      params.denoising.use = false;
-    }
-  }
-
   /* Viewport Performance */
-  params.start_resolution = get_int(cscene, "preview_start_resolution");
   params.pixel_size = b_engine.get_preview_pixel_size(b_scene);
 
-  /* other parameters */
-  params.cancel_timeout = (double)get_float(cscene, "debug_cancel_timeout");
-  params.reset_timeout = (double)get_float(cscene, "debug_reset_timeout");
-  params.text_timeout = (double)get_float(cscene, "debug_text_timeout");
-
-  /* progressive refine */
-  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)
-    params.progressive_refine = false;
-
   if (background) {
-    if (params.progressive_refine)
-      params.progressive = true;
-    else
-      params.progressive = false;
-
-    params.start_resolution = INT_MAX;
     params.pixel_size = 1;
   }
-  else
-    params.progressive = true;
 
   /* shading system - scene level needs full refresh */
   const bool shadingsystem = RNA_boolean_get(&cscene, "shading_system");
@@ -950,19 +837,30 @@ SessionParams BlenderSync::get_session_params(BL::RenderEngine &b_engine,
   else if (shadingsystem == 1)
     params.shadingsystem = SHADINGSYSTEM_OSL;
 
-  /* Color management. */
-  params.display_buffer_linear = b_engine.support_display_space_shader(b_scene);
-
-  if (b_engine.is_preview()) {
-    /* For preview rendering we're using same timeout as
-     * blender's job update.
-     */
-    params.progressive_update_timeout = 0.1;
+  /* Time limit. */
+  if (background) {
+    params.time_limit = get_float(cscene, "time_limit");
+  }
+  else {
+    /* For the viewport it kind of makes more sense to think in terms of the noise floor, which is
+     * usually higher than acceptable level for the final frame. */
+    /* TODO: It might be useful to support time limit in the viewport as well, but needs some
+     * extra thoughts and input. */
+    params.time_limit = 0.0;
   }
 
+  /* Profiling. */
   params.use_profiling = params.device.has_profiling && !b_engine.is_preview() && background &&
                          BlenderSession::print_render_stats;
 
+  if (background) {
+    params.use_auto_tile = RNA_boolean_get(&cscene, "use_auto_tile");
+    params.tile_size = max(get_int(cscene, "tile_size"), 8);
+  }
+  else {
+    params.use_auto_tile = false;
+  }
+
   return params;
 }
 
@@ -970,33 +868,34 @@ DenoiseParams BlenderSync::get_denoise_params(BL::Scene &b_scene,
                                               BL::ViewLayer &b_view_layer,
                                               bool background)
 {
+  enum DenoiserInput {
+    DENOISER_INPUT_RGB = 1,
+    DENOISER_INPUT_RGB_ALBEDO = 2,
+    DENOISER_INPUT_RGB_ALBEDO_NORMAL = 3,
+
+    DENOISER_INPUT_NUM,
+  };
+
   DenoiseParams denoising;
   PointerRNA cscene = RNA_pointer_get(&b_scene.ptr, "cycles");
 
+  int input_passes = -1;
+
   if (background) {
     /* Final Render Denoising */
     denoising.use = get_boolean(cscene, "use_denoising");
     denoising.type = (DenoiserType)get_enum(cscene, "denoiser", DENOISER_NUM, DENOISER_NONE);
+    denoising.prefilter = (DenoiserPrefilter)get_enum(
+        cscene, "denoising_prefilter", DENOISER_PREFILTER_NUM, DENOISER_PREFILTER_NONE);
+
+    input_passes = (DenoiserInput)get_enum(
+        cscene, "denoising_input_passes", DENOISER_INPUT_NUM, DENOISER_INPUT_RGB_ALBEDO_NORMAL);
 
     if (b_view_layer) {
       PointerRNA clayer = RNA_pointer_get(&b_view_layer.ptr, "cycles");
       if (!get_boolean(clayer, "use_denoising")) {
         denoising.use = false;
       }
-
-      denoising.radius = get_int(clayer, "denoising_radius");
-      denoising.strength = get_float(clayer, "denoising_strength");
-      denoising.feature_strength = get_float(clayer, "denoising_feature_strength");
-      denoising.relative_pca = get_boolean(clayer, "denoising_relative_pca");
-
-      denoising.input_passes = (DenoiserInput)get_enum(
-          clayer,
-          (denoising.type == DENOISER_OPTIX) ? "denoising_optix_input_passes" :
-                                               "denoising_openimagedenoise_input_passes",
-          DENOISER_INPUT_NUM,
-          DENOISER_INPUT_RGB_ALBEDO_NORMAL);
-
-      denoising.store_passes = get_boolean(clayer, "denoising_store_passes");
     }
   }
   else {
@@ -1004,10 +903,12 @@ DenoiseParams BlenderSync::get_denoise_params(BL::Scene &b_scene,
     denoising.use = get_boolean(cscene, "use_preview_denoising");
     denoising.type = (DenoiserType)get_enum(
         cscene, "preview_denoiser", DENOISER_NUM, DENOISER_NONE);
+    denoising.prefilter = (DenoiserPrefilter)get_enum(
+        cscene, "preview_denoising_prefilter", DENOISER_PREFILTER_NUM, DENOISER_PREFILTER_FAST);
     denoising.start_sample = get_int(cscene, "preview_denoising_start_sample");
 
-    denoising.input_passes = (DenoiserInput)get_enum(
-        cscene, "preview_denoising_input_passes", DENOISER_INPUT_NUM, (int)denoising.input_passes);
+    input_passes = (DenoiserInput)get_enum(
+        cscene, "preview_denoising_input_passes", DENOISER_INPUT_NUM, DENOISER_INPUT_RGB_ALBEDO);
 
     /* Auto select fastest denoiser. */
     if (denoising.type == DENOISER_NONE) {
@@ -1023,6 +924,27 @@ DenoiseParams BlenderSync::get_denoise_params(BL::Scene &b_scene,
     }
   }
 
+  switch (input_passes) {
+    case DENOISER_INPUT_RGB:
+      denoising.use_pass_albedo = false;
+      denoising.use_pass_normal = false;
+      break;
+
+    case DENOISER_INPUT_RGB_ALBEDO:
+      denoising.use_pass_albedo = true;
+      denoising.use_pass_normal = false;
+      break;
+
+    case DENOISER_INPUT_RGB_ALBEDO_NORMAL:
+      denoising.use_pass_albedo = true;
+      denoising.use_pass_normal = true;
+      break;
+
+    default:
+      LOG(ERROR) << "Unhandled input passes enum " << input_passes;
+      break;
+  }
+
   return denoising;
 }
 

intern/cycles/blender/blender_sync.h

@@ -60,6 +60,7 @@ class BlenderSync {
               BL::Scene &b_scene,
               Scene *scene,
               bool preview,
+              bool use_developer_ui,
               Progress &progress);
   ~BlenderSync();
 
@@ -75,12 +76,8 @@ class BlenderSync {
                  int height,
                  void **python_thread_state);
   void sync_view_layer(BL::ViewLayer &b_view_layer);
-  vector<Pass> sync_render_passes(BL::Scene &b_scene,
-                                  BL::RenderLayer &b_render_layer,
-                                  BL::ViewLayer &b_view_layer,
-                                  bool adaptive_sampling,
-                                  const DenoiseParams &denoising);
-  void sync_integrator();
+  void sync_render_passes(BL::RenderLayer &b_render_layer, BL::ViewLayer &b_view_layer);
+  void sync_integrator(BL::ViewLayer &b_view_layer, bool background);
   void sync_camera(BL::RenderSettings &b_render,
                    BL::Object &b_override,
                    int width,
@@ -98,22 +95,13 @@ class BlenderSync {
 
   /* get parameters */
   static SceneParams get_scene_params(BL::Scene &b_scene, bool background);
-  static SessionParams get_session_params(
-      BL::RenderEngine &b_engine,
-      BL::Preferences &b_userpref,
-      BL::Scene &b_scene,
-      bool background,
-      BL::ViewLayer b_view_layer = BL::ViewLayer(PointerRNA_NULL));
+  static SessionParams get_session_params(BL::RenderEngine &b_engine,
+                                          BL::Preferences &b_userpref,
+                                          BL::Scene &b_scene,
+                                          bool background);
   static bool get_session_pause(BL::Scene &b_scene, bool background);
-  static BufferParams get_buffer_params(BL::SpaceView3D &b_v3d,
-                                        BL::RegionView3D &b_rv3d,
-                                        Camera *cam,
-                                        int width,
-                                        int height,
-                                        const bool use_denoiser);
-
-  static PassType get_pass_type(BL::RenderPass &b_pass);
-  static int get_denoising_pass(BL::RenderPass &b_pass);
+  static BufferParams get_buffer_params(
+      BL::SpaceView3D &b_v3d, BL::RegionView3D &b_rv3d, Camera *cam, int width, int height);
 
  private:
   static DenoiseParams get_denoise_params(BL::Scene &b_scene,
@@ -131,7 +119,7 @@ class BlenderSync {
                    int width,
                    int height,
                    void **python_thread_state);
-  void sync_film(BL::SpaceView3D &b_v3d);
+  void sync_film(BL::ViewLayer &b_view_layer, BL::SpaceView3D &b_v3d);
   void sync_view();
 
   /* Shader */
@@ -245,6 +233,7 @@ class BlenderSync {
   Scene *scene;
   bool preview;
   bool experimental;
+  bool use_developer_ui;
 
   float dicing_rate;
   int max_subdivisions;
@@ -253,7 +242,6 @@ class BlenderSync {
     RenderLayerInfo()
         : material_override(PointerRNA_NULL),
           use_background_shader(true),
-          use_background_ao(true),
           use_surfaces(true),
           use_hair(true),
           use_volumes(true),
@@ -266,7 +254,6 @@ class BlenderSync {
     string name;
     BL::Material material_override;
     bool use_background_shader;
-    bool use_background_ao;
     bool use_surfaces;
     bool use_hair;
     bool use_volumes;

intern/cycles/blender/blender_util.h

@@ -90,26 +90,27 @@ static inline BL::Mesh object_to_mesh(BL::BlendData & /*data*/,
   }
 #endif
 
-  BL::Mesh mesh(PointerRNA_NULL);
-  if (b_ob_info.object_data.is_a(&RNA_Mesh)) {
-    /* TODO: calc_undeformed is not used. */
-    mesh = BL::Mesh(b_ob_info.object_data);
+  BL::Mesh mesh = (b_ob_info.object_data.is_a(&RNA_Mesh)) ? BL::Mesh(b_ob_info.object_data) :
+                                                            BL::Mesh(PointerRNA_NULL);
 
-    /* Make a copy to split faces if we use autosmooth, otherwise not needed.
-     * Also in edit mode do we need to make a copy, to ensure data layers like
-     * UV are not empty. */
-    if (mesh.is_editmode() ||
-        (mesh.use_auto_smooth() && subdivision_type == Mesh::SUBDIVISION_NONE)) {
+  if (b_ob_info.is_real_object_data()) {
+    if (mesh) {
+      /* Make a copy to split faces if we use autosmooth, otherwise not needed.
+       * Also in edit mode do we need to make a copy, to ensure data layers like
+       * UV are not empty. */
+      if (mesh.is_editmode() ||
+          (mesh.use_auto_smooth() && subdivision_type == Mesh::SUBDIVISION_NONE)) {
+        BL::Depsgraph depsgraph(PointerRNA_NULL);
+        mesh = b_ob_info.real_object.to_mesh(false, depsgraph);
+      }
+    }
+    else {
       BL::Depsgraph depsgraph(PointerRNA_NULL);
-      assert(b_ob_info.is_real_object_data());
       mesh = b_ob_info.real_object.to_mesh(false, depsgraph);
     }
   }
   else {
-    BL::Depsgraph depsgraph(PointerRNA_NULL);
-    if (b_ob_info.is_real_object_data()) {
-      mesh = b_ob_info.real_object.to_mesh(false, depsgraph);
-    }
+    /* TODO: what to do about non-mesh geometry instances? */
   }
 
 #if 0

intern/cycles/blender/blender_viewport.cpp

@@ -17,6 +17,8 @@
 #include "blender_viewport.h"
 
 #include "blender_util.h"
+#include "render/pass.h"
+#include "util/util_logging.h"
 
 CCL_NAMESPACE_BEGIN
 
@@ -26,11 +28,12 @@ BlenderViewportParameters::BlenderViewportParameters()
       studiolight_rotate_z(0.0f),
       studiolight_intensity(1.0f),
       studiolight_background_alpha(1.0f),
-      display_pass(PASS_COMBINED)
+      display_pass(PASS_COMBINED),
+      show_active_pixels(false)
 {
 }
 
-BlenderViewportParameters::BlenderViewportParameters(BL::SpaceView3D &b_v3d)
+BlenderViewportParameters::BlenderViewportParameters(BL::SpaceView3D &b_v3d, bool use_developer_ui)
     : BlenderViewportParameters()
 {
   if (!b_v3d) {
@@ -55,7 +58,25 @@ BlenderViewportParameters::BlenderViewportParameters(BL::SpaceView3D &b_v3d)
   }
 
   /* Film. */
-  display_pass = (PassType)get_enum(cshading, "render_pass", -1, -1);
+
+  /* Lookup display pass based on the enum identifier.
+   * This is because integer values of python enum are not aligned with the passes definition in
+   * the kernel. */
+
+  display_pass = PASS_COMBINED;
+
+  const string display_pass_identifier = get_enum_identifier(cshading, "render_pass");
+  if (!display_pass_identifier.empty()) {
+    const ustring pass_type_identifier(string_to_lower(display_pass_identifier));
+    const NodeEnum *pass_type_enum = Pass::get_type_enum();
+    if (pass_type_enum->exists(pass_type_identifier)) {
+      display_pass = static_cast<PassType>((*pass_type_enum)[pass_type_identifier]);
+    }
+  }
+
+  if (use_developer_ui) {
+    show_active_pixels = get_boolean(cshading, "show_active_pixels");
+  }
 }
 
 bool BlenderViewportParameters::shader_modified(const BlenderViewportParameters &other) const
@@ -69,7 +90,7 @@ bool BlenderViewportParameters::shader_modified(const BlenderViewportParameters
 
 bool BlenderViewportParameters::film_modified(const BlenderViewportParameters &other) const
 {
-  return display_pass != other.display_pass;
+  return display_pass != other.display_pass || show_active_pixels != other.show_active_pixels;
 }
 
 bool BlenderViewportParameters::modified(const BlenderViewportParameters &other) const
@@ -82,18 +103,4 @@ bool BlenderViewportParameters::use_custom_shader() const
   return !(use_scene_world && use_scene_lights);
 }
 
-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;
-
-    passes.clear();
-    Pass::add(display_pass, passes);
-
-    return display_pass;
-  }
-  return PASS_NONE;
-}
-
 CCL_NAMESPACE_END

intern/cycles/blender/blender_viewport.h

@@ -39,9 +39,10 @@ class BlenderViewportParameters {
 
   /* Film. */
   PassType display_pass;
+  bool show_active_pixels;
 
   BlenderViewportParameters();
-  explicit BlenderViewportParameters(BL::SpaceView3D &b_v3d);
+  BlenderViewportParameters(BL::SpaceView3D &b_v3d, bool use_developer_ui);
 
   /* Check whether any of shading related settings are different from the given parameters. */
   bool shader_modified(const BlenderViewportParameters &other) const;
@@ -57,8 +58,6 @@ class BlenderViewportParameters {
   bool use_custom_shader() const;
 };
 
-PassType update_viewport_display_passes(BL::SpaceView3D &b_v3d, vector<Pass> &passes);
-
 CCL_NAMESPACE_END
 
 #endif

intern/cycles/bvh/bvh_build.cpp

@@ -832,18 +832,18 @@ BVHNode *BVHBuild::create_leaf_node(const BVHRange &range, const vector<BVHRefer
   typedef StackAllocator<256, float2> LeafTimeStackAllocator;
   typedef StackAllocator<256, BVHReference> LeafReferenceStackAllocator;
 
-  vector<int, LeafStackAllocator> p_type[PRIMITIVE_NUM_TOTAL];
-  vector<int, LeafStackAllocator> p_index[PRIMITIVE_NUM_TOTAL];
-  vector<int, LeafStackAllocator> p_object[PRIMITIVE_NUM_TOTAL];
-  vector<float2, LeafTimeStackAllocator> p_time[PRIMITIVE_NUM_TOTAL];
-  vector<BVHReference, LeafReferenceStackAllocator> p_ref[PRIMITIVE_NUM_TOTAL];
+  vector<int, LeafStackAllocator> p_type[PRIMITIVE_NUM];
+  vector<int, LeafStackAllocator> p_index[PRIMITIVE_NUM];
+  vector<int, LeafStackAllocator> p_object[PRIMITIVE_NUM];
+  vector<float2, LeafTimeStackAllocator> p_time[PRIMITIVE_NUM];
+  vector<BVHReference, LeafReferenceStackAllocator> p_ref[PRIMITIVE_NUM];
 
   /* TODO(sergey): In theory we should be able to store references. */
   vector<BVHReference, LeafReferenceStackAllocator> object_references;
 
-  uint visibility[PRIMITIVE_NUM_TOTAL] = {0};
+  uint visibility[PRIMITIVE_NUM] = {0};
   /* NOTE: Keep initialization in sync with actual number of primitives. */
-  BoundBox bounds[PRIMITIVE_NUM_TOTAL] = {
+  BoundBox bounds[PRIMITIVE_NUM] = {
       BoundBox::empty, BoundBox::empty, BoundBox::empty, BoundBox::empty};
   int ob_num = 0;
   int num_new_prims = 0;
@@ -877,7 +877,7 @@ BVHNode *BVHBuild::create_leaf_node(const BVHRange &range, const vector<BVHRefer
    * TODO(sergey): With some pointer trickery we can write directly to the
    * destination buffers for the non-spatial split BVH.
    */
-  BVHNode *leaves[PRIMITIVE_NUM_TOTAL + 1] = {NULL};
+  BVHNode *leaves[PRIMITIVE_NUM + 1] = {NULL};
   int num_leaves = 0;
   size_t start_index = 0;
   vector<int, LeafStackAllocator> local_prim_type, local_prim_index, local_prim_object;
@@ -888,7 +888,7 @@ BVHNode *BVHBuild::create_leaf_node(const BVHRange &range, const vector<BVHRefer
   if (need_prim_time) {
     local_prim_time.resize(num_new_prims);
   }
-  for (int i = 0; i < PRIMITIVE_NUM_TOTAL; ++i) {
+  for (int i = 0; i < PRIMITIVE_NUM; ++i) {
     int num = (int)p_type[i].size();
     if (num != 0) {
       assert(p_type[i].size() == p_index[i].size());

intern/cycles/bvh/bvh_embree.cpp

@@ -37,10 +37,10 @@
 /* Kernel includes are necessary so that the filter function for Embree can access the packed BVH.
  */
 #  include "kernel/bvh/bvh_embree.h"
-#  include "kernel/kernel_compat_cpu.h"
-#  include "kernel/kernel_globals.h"
+#  include "kernel/bvh/bvh_util.h"
+#  include "kernel/device/cpu/compat.h"
+#  include "kernel/device/cpu/globals.h"
 #  include "kernel/kernel_random.h"
-#  include "kernel/split/kernel_split_data_types.h"
 
 #  include "render/hair.h"
 #  include "render/mesh.h"
@@ -73,46 +73,69 @@ static void rtc_filter_occluded_func(const RTCFilterFunctionNArguments *args)
   const RTCRay *ray = (RTCRay *)args->ray;
   RTCHit *hit = (RTCHit *)args->hit;
   CCLIntersectContext *ctx = ((IntersectContext *)args->context)->userRayExt;
-  KernelGlobals *kg = ctx->kg;
+  const KernelGlobals *kg = ctx->kg;
 
   switch (ctx->type) {
     case CCLIntersectContext::RAY_SHADOW_ALL: {
-      /* Append the intersection to the end of the array. */
-      if (ctx->num_hits < ctx->max_hits) {
-        Intersection current_isect;
-        kernel_embree_convert_hit(kg, ray, hit, &current_isect);
-        for (size_t i = 0; i < ctx->max_hits; ++i) {
+      Intersection current_isect;
+      kernel_embree_convert_hit(kg, ray, hit, &current_isect);
+
+      /* If no transparent shadows, all light is blocked. */
+      const int flags = intersection_get_shader_flags(kg, &current_isect);
+      if (!(flags & (SD_HAS_TRANSPARENT_SHADOW)) || ctx->max_hits == 0) {
+        ctx->opaque_hit = true;
+        return;
+      }
+
+      /* Test if we need to record this transparent intersection. */
+      if (ctx->num_hits < ctx->max_hits || ray->tfar < ctx->max_t) {
+        /* Skip already recorded intersections. */
+        int num_recorded_hits = min(ctx->num_hits, ctx->max_hits);
+
+        for (int i = 0; i < num_recorded_hits; ++i) {
           if (current_isect.object == ctx->isect_s[i].object &&
               current_isect.prim == ctx->isect_s[i].prim && current_isect.t == ctx->isect_s[i].t) {
             /* This intersection was already recorded, skip it. */
             *args->valid = 0;
-            break;
+            return;
           }
         }
-        Intersection *isect = &ctx->isect_s[ctx->num_hits];
-        ++ctx->num_hits;
-        *isect = current_isect;
-        int prim = kernel_tex_fetch(__prim_index, isect->prim);
-        int shader = 0;
-        if (kernel_tex_fetch(__prim_type, isect->prim) & PRIMITIVE_ALL_TRIANGLE) {
-          shader = kernel_tex_fetch(__tri_shader, prim);
-        }
-        else {
-          float4 str = kernel_tex_fetch(__curves, prim);
-          shader = __float_as_int(str.z);
-        }
-        int flag = kernel_tex_fetch(__shaders, shader & SHADER_MASK).flags;
-        /* If no transparent shadows, all light is blocked. */
-        if (flag & (SD_HAS_TRANSPARENT_SHADOW)) {
-          /* This tells Embree to continue tracing. */
-          *args->valid = 0;
+
+        /* If maximum number of hits was reached, replace the intersection with the
+         * highest distance. We want to find the N closest intersections. */
+        int isect_index = num_recorded_hits;
+        if (num_recorded_hits + 1 >= ctx->max_hits) {
+          float max_t = ctx->isect_s[0].t;
+          int max_recorded_hit = 0;
+
+          for (int i = 1; i < num_recorded_hits; ++i) {
+            if (ctx->isect_s[i].t > max_t) {
+              max_recorded_hit = i;
+              max_t = ctx->isect_s[i].t;
+            }
+          }
+
+          if (num_recorded_hits >= ctx->max_hits) {
+            isect_index = max_recorded_hit;
+          }
+
+          /* Limit the ray distance and stop counting hits beyond this.
+           * TODO: is there some way we can tell Embree to stop intersecting beyond
+           * this distance when max number of hits is reached?. Or maybe it will
+           * become irrelevant if we make max_hits a very high number on the CPU. */
+          ctx->max_t = max(current_isect.t, max_t);
         }
+
+        ctx->isect_s[isect_index] = current_isect;
       }
-      else {
-        /* Increase the number of hits beyond ray.max_hits
-         * so that the caller can detect this as opaque. */
-        ++ctx->num_hits;
-      }
+
+      /* Always increase the number of hits, even beyond ray.max_hits so that
+       * the caller can detect this as and consider it opaque, or trace another
+       * ray. */
+      ++ctx->num_hits;
+
+      /* This tells Embree to continue tracing. */
+      *args->valid = 0;
       break;
     }
     case CCLIntersectContext::RAY_LOCAL:
@@ -329,7 +352,7 @@ void BVHEmbree::build(Progress &progress, Stats *stats, RTCDevice rtc_device_)
     scene = NULL;
   }
 
-  const bool dynamic = params.bvh_type == SceneParams::BVH_DYNAMIC;
+  const bool dynamic = params.bvh_type == BVH_TYPE_DYNAMIC;
 
   scene = rtcNewScene(rtc_device);
   const RTCSceneFlags scene_flags = (dynamic ? RTC_SCENE_FLAG_DYNAMIC : RTC_SCENE_FLAG_NONE) |

intern/cycles/bvh/bvh_params.h

@@ -31,6 +31,27 @@ CCL_NAMESPACE_BEGIN
  */
 typedef KernelBVHLayout BVHLayout;
 
+/* Type of BVH, in terms whether it is supported dynamic updates of meshes
+ * or whether modifying geometry requires full BVH rebuild.
+ */
+enum BVHType {
+  /* BVH supports dynamic updates of geometry.
+   *
+   * Faster for updating BVH tree when doing modifications in viewport,
+   * but slower for rendering.
+   */
+  BVH_TYPE_DYNAMIC = 0,
+  /* BVH tree is calculated for specific scene, updates in geometry
+   * requires full tree rebuild.
+   *
+   * Slower to update BVH tree when modifying objects in viewport, also
+   * slower to build final BVH tree but gives best possible render speed.
+   */
+  BVH_TYPE_STATIC = 1,
+
+  BVH_NUM_TYPES,
+};
+
 /* Names bitflag type to denote which BVH layouts are supported by
  * particular area.
  *

intern/cycles/cmake/external_libs.cmake

@@ -287,9 +287,6 @@ if(CYCLES_STANDALONE_REPOSITORY)
   endif()
 
   set(__boost_packages filesystem regex system thread date_time)
-  if(WITH_CYCLES_NETWORK)
-    list(APPEND __boost_packages serialization)
-  endif()
   if(WITH_CYCLES_OSL)
     list(APPEND __boost_packages wave)
   endif()
@@ -535,4 +532,13 @@ if(WITH_CYCLES_CUDA_BINARIES OR NOT WITH_CUDA_DYNLOAD)
   endif()
 endif()
 
+
+###########################################################################
+# HIP
+###########################################################################
+
+if(NOT WITH_HIP_DYNLOAD)
+  set(WITH_HIP_DYNLOAD ON)
+endif()
+
 unset(_cycles_lib_dir)

intern/cycles/cmake/macros.cmake

@@ -156,10 +156,16 @@ macro(cycles_target_link_libraries target)
     ${PLATFORM_LINKLIBS}
   )
 
-  if(WITH_CUDA_DYNLOAD)
-    target_link_libraries(${target} extern_cuew)
-  else()
-    target_link_libraries(${target} ${CUDA_CUDA_LIBRARY})
+  if(WITH_CYCLES_DEVICE_CUDA OR WITH_CYCLES_DEVICE_OPTIX)
+    if(WITH_CUDA_DYNLOAD)
+      target_link_libraries(${target} extern_cuew)
+    else()
+      target_link_libraries(${target} ${CUDA_CUDA_LIBRARY})
+    endif()
+  endif()
+
+  if(WITH_CYCLES_DEVICE_HIP AND WITH_HIP_DYNLOAD)
+    target_link_libraries(${target} extern_hipew)
   endif()
 
   if(CYCLES_STANDALONE_REPOSITORY)

intern/cycles/device/CMakeLists.txt

@@ -22,91 +22,139 @@ set(INC_SYS
   ../../../extern/clew/include
 )
 
-if(WITH_CUDA_DYNLOAD)
+if(WITH_CYCLES_DEVICE_OPTIX OR WITH_CYCLES_DEVICE_CUDA)
+  if(WITH_CUDA_DYNLOAD)
+    list(APPEND INC
+      ../../../extern/cuew/include
+    )
+    add_definitions(-DWITH_CUDA_DYNLOAD)
+  else()
+    list(APPEND INC_SYS
+      ${CUDA_TOOLKIT_INCLUDE}
+    )
+    add_definitions(-DCYCLES_CUDA_NVCC_EXECUTABLE="${CUDA_NVCC_EXECUTABLE}")
+  endif()
+endif()
+
+if(WITH_CYCLES_DEVICE_HIP AND WITH_HIP_DYNLOAD)
   list(APPEND INC
-    ../../../extern/cuew/include
+    ../../../extern/hipew/include
   )
-  add_definitions(-DWITH_CUDA_DYNLOAD)
-else()
-  list(APPEND INC_SYS
-    ${CUDA_TOOLKIT_INCLUDE}
-  )
-  add_definitions(-DCYCLES_CUDA_NVCC_EXECUTABLE="${CUDA_NVCC_EXECUTABLE}")
+  add_definitions(-DWITH_HIP_DYNLOAD)
 endif()
 
 set(SRC
   device.cpp
-  device_cpu.cpp
-  device_cuda.cpp
-  device_denoising.cpp
-  device_dummy.cpp
+  device_denoise.cpp
+  device_graphics_interop.cpp
+  device_kernel.cpp
   device_memory.cpp
-  device_multi.cpp
-  device_opencl.cpp
-  device_optix.cpp
-  device_split_kernel.cpp
-  device_task.cpp
+  device_queue.cpp
+)
+
+set(SRC_CPU
+  cpu/device.cpp
+  cpu/device.h
+  cpu/device_impl.cpp
+  cpu/device_impl.h
+  cpu/kernel.cpp
+  cpu/kernel.h
+  cpu/kernel_function.h
+  cpu/kernel_thread_globals.cpp
+  cpu/kernel_thread_globals.h
 )
 
 set(SRC_CUDA
-  cuda/device_cuda.h
-  cuda/device_cuda_impl.cpp
+  cuda/device.cpp
+  cuda/device.h
+  cuda/device_impl.cpp
+  cuda/device_impl.h
+  cuda/graphics_interop.cpp
+  cuda/graphics_interop.h
+  cuda/kernel.cpp
+  cuda/kernel.h
+  cuda/queue.cpp
+  cuda/queue.h
+  cuda/util.cpp
+  cuda/util.h
 )
 
-set(SRC_OPENCL
-  opencl/device_opencl.h
-  opencl/device_opencl_impl.cpp
-  opencl/memory_manager.h
-  opencl/memory_manager.cpp
-  opencl/opencl_util.cpp
+set(SRC_HIP
+  hip/device.cpp
+  hip/device.h
+  hip/device_impl.cpp
+  hip/device_impl.h
+  hip/graphics_interop.cpp
+  hip/graphics_interop.h
+  hip/kernel.cpp
+  hip/kernel.h
+  hip/queue.cpp
+  hip/queue.h
+  hip/util.cpp
+  hip/util.h
 )
 
-if(WITH_CYCLES_NETWORK)
-  list(APPEND SRC
-    device_network.cpp
-  )
-endif()
+set(SRC_DUMMY
+  dummy/device.cpp
+  dummy/device.h
+)
+
+set(SRC_MULTI
+  multi/device.cpp
+  multi/device.h
+)
+
+set(SRC_OPTIX
+  optix/device.cpp
+  optix/device.h
+  optix/device_impl.cpp
+  optix/device_impl.h
+  optix/queue.cpp
+  optix/queue.h
+  optix/util.h
+)
 
 set(SRC_HEADERS
   device.h
-  device_denoising.h
+  device_denoise.h
+  device_graphics_interop.h
   device_memory.h
-  device_intern.h
-  device_network.h
-  device_split_kernel.h
-  device_task.h
+  device_kernel.h
+  device_queue.h
 )
 
 set(LIB
-  cycles_render
   cycles_kernel
   cycles_util
   ${CYCLES_GL_LIBRARIES}
 )
 
-if(WITH_CUDA_DYNLOAD)
+if(WITH_CYCLES_DEVICE_OPTIX OR WITH_CYCLES_DEVICE_CUDA)
+  if(WITH_CUDA_DYNLOAD)
+    list(APPEND LIB
+      extern_cuew
+    )
+  else()
+    list(APPEND LIB
+      ${CUDA_CUDA_LIBRARY}
+    )
+  endif()
+endif()
+
+if(WITH_CYCLES_DEVICE_HIP AND WITH_HIP_DYNLOAD)
   list(APPEND LIB
-    extern_cuew
-  )
-else()
-  list(APPEND LIB
-    ${CUDA_CUDA_LIBRARY}
+    extern_hipew
   )
 endif()
 
 add_definitions(${GL_DEFINITIONS})
-if(WITH_CYCLES_NETWORK)
-  add_definitions(-DWITH_NETWORK)
-endif()
-if(WITH_CYCLES_DEVICE_OPENCL)
-  list(APPEND LIB
-    extern_clew
-  )
-  add_definitions(-DWITH_OPENCL)
-endif()
+
 if(WITH_CYCLES_DEVICE_CUDA)
   add_definitions(-DWITH_CUDA)
 endif()
+if(WITH_CYCLES_DEVICE_HIP)
+  add_definitions(-DWITH_HIP)
+endif()
 if(WITH_CYCLES_DEVICE_OPTIX)
   add_definitions(-DWITH_OPTIX)
 endif()
@@ -115,18 +163,28 @@ if(WITH_CYCLES_DEVICE_MULTI)
 endif()
 
 if(WITH_OPENIMAGEDENOISE)
-  add_definitions(-DWITH_OPENIMAGEDENOISE)
-  add_definitions(-DOIDN_STATIC_LIB)
-  list(APPEND INC_SYS
-    ${OPENIMAGEDENOISE_INCLUDE_DIRS}
-  )
   list(APPEND LIB
     ${OPENIMAGEDENOISE_LIBRARIES}
-    ${TBB_LIBRARIES}
   )
 endif()
 
 include_directories(${INC})
 include_directories(SYSTEM ${INC_SYS})
 
-cycles_add_library(cycles_device "${LIB}" ${SRC} ${SRC_CUDA} ${SRC_OPENCL} ${SRC_HEADERS})
+cycles_add_library(cycles_device "${LIB}"
+  ${SRC}
+  ${SRC_CPU}
+  ${SRC_CUDA}
+  ${SRC_HIP}
+  ${SRC_DUMMY}
+  ${SRC_MULTI}
+  ${SRC_OPTIX}
+  ${SRC_HEADERS}
+)
+
+source_group("cpu" FILES ${SRC_CPU})
+source_group("cuda" FILES ${SRC_CUDA})
+source_group("dummy" FILES ${SRC_DUMMY})
+source_group("multi" FILES ${SRC_MULTI})
+source_group("optix" FILES ${SRC_OPTIX})
+source_group("common" FILES ${SRC} ${SRC_HEADERS})

intern/cycles/device/cpu/device.cpp

@@ -0,0 +1,64 @@
+/*
+ * Copyright 2011-2013 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.
+ */
+
+#include "device/cpu/device.h"
+#include "device/cpu/device_impl.h"
+
+/* Used for `info.denoisers`. */
+/* TODO(sergey): The denoisers are probably to be moved completely out of the device into their
+ * own class. But until then keep API consistent with how it used to work before. */
+#include "util/util_openimagedenoise.h"
+
+CCL_NAMESPACE_BEGIN
+
+Device *device_cpu_create(const DeviceInfo &info, Stats &stats, Profiler &profiler)
+{
+  return new CPUDevice(info, stats, profiler);
+}
+
+void device_cpu_info(vector<DeviceInfo> &devices)
+{
+  DeviceInfo info;
+
+  info.type = DEVICE_CPU;
+  info.description = system_cpu_brand_string();
+  info.id = "CPU";
+  info.num = 0;
+  info.has_osl = true;
+  info.has_half_images = true;
+  info.has_nanovdb = true;
+  info.has_profiling = true;
+  if (openimagedenoise_supported()) {
+    info.denoisers |= DENOISER_OPENIMAGEDENOISE;
+  }
+
+  devices.insert(devices.begin(), info);
+}
+
+string device_cpu_capabilities()
+{
+  string capabilities = "";
+  capabilities += system_cpu_support_sse2() ? "SSE2 " : "";
+  capabilities += system_cpu_support_sse3() ? "SSE3 " : "";
+  capabilities += system_cpu_support_sse41() ? "SSE41 " : "";
+  capabilities += system_cpu_support_avx() ? "AVX " : "";
+  capabilities += system_cpu_support_avx2() ? "AVX2" : "";
+  if (capabilities[capabilities.size() - 1] == ' ')
+    capabilities.resize(capabilities.size() - 1);
+  return capabilities;
+}
+
+CCL_NAMESPACE_END

intern/cycles/device/cpu/device.h

@@ -1,5 +1,5 @@
 /*
- * Copyright 2011-2015 Blender Foundation
+ * Copyright 2011-2021 Blender Foundation
  *
  * Licensed under the Apache License, Version 2.0 (the "License");
  * you may not use this file except in compliance with the License.
@@ -14,13 +14,22 @@
  * limitations under the License.
  */
 
-#include "kernel/kernel_compat_opencl.h"
-#include "kernel/split/kernel_split_common.h"
-#include "kernel/split/kernel_buffer_update.h"
+#pragma once
 
-#define KERNEL_NAME buffer_update
-#define LOCALS_TYPE unsigned int
-#include "kernel/kernels/opencl/kernel_split_function.h"
-#undef KERNEL_NAME
-#undef LOCALS_TYPE
+#include "util/util_string.h"
+#include "util/util_vector.h"
 
+CCL_NAMESPACE_BEGIN
+
+class Device;
+class DeviceInfo;
+class Profiler;
+class Stats;
+
+Device *device_cpu_create(const DeviceInfo &info, Stats &stats, Profiler &profiler);
+
+void device_cpu_info(vector<DeviceInfo> &devices);
+
+string device_cpu_capabilities();
+
+CCL_NAMESPACE_END

intern/cycles/device/cpu/device_impl.cpp

@@ -0,0 +1,332 @@
+/*
+ * Copyright 2011-2021 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.
+ */
+
+#include "device/cpu/device_impl.h"
+
+#include <stdlib.h>
+#include <string.h>
+
+/* So ImathMath is included before our kernel_cpu_compat. */
+#ifdef WITH_OSL
+/* So no context pollution happens from indirectly included windows.h */
+#  include "util/util_windows.h"
+#  include <OSL/oslexec.h>
+#endif
+
+#ifdef WITH_EMBREE
+#  include <embree3/rtcore.h>
+#endif
+
+#include "device/cpu/kernel.h"
+#include "device/cpu/kernel_thread_globals.h"
+
+#include "device/device.h"
+
+// clang-format off
+#include "kernel/device/cpu/compat.h"
+#include "kernel/device/cpu/globals.h"
+#include "kernel/device/cpu/kernel.h"
+#include "kernel/kernel_types.h"
+
+#include "kernel/osl/osl_shader.h"
+#include "kernel/osl/osl_globals.h"
+// clang-format on
+
+#include "bvh/bvh_embree.h"
+
+#include "render/buffers.h"
+
+#include "util/util_debug.h"
+#include "util/util_foreach.h"
+#include "util/util_function.h"
+#include "util/util_logging.h"
+#include "util/util_map.h"
+#include "util/util_openimagedenoise.h"
+#include "util/util_optimization.h"
+#include "util/util_progress.h"
+#include "util/util_system.h"
+#include "util/util_task.h"
+#include "util/util_thread.h"
+
+CCL_NAMESPACE_BEGIN
+
+CPUDevice::CPUDevice(const DeviceInfo &info_, Stats &stats_, Profiler &profiler_)
+    : Device(info_, stats_, profiler_), texture_info(this, "__texture_info", MEM_GLOBAL)
+{
+  /* Pick any kernel, all of them are supposed to have same level of microarchitecture
+   * optimization. */
+  VLOG(1) << "Will be using " << kernels.integrator_init_from_camera.get_uarch_name()
+          << " kernels.";
+
+  if (info.cpu_threads == 0) {
+    info.cpu_threads = TaskScheduler::num_threads();
+  }
+
+#ifdef WITH_OSL
+  kernel_globals.osl = &osl_globals;
+#endif
+#ifdef WITH_EMBREE
+  embree_device = rtcNewDevice("verbose=0");
+#endif
+  need_texture_info = false;
+}
+
+CPUDevice::~CPUDevice()
+{
+#ifdef WITH_EMBREE
+  rtcReleaseDevice(embree_device);
+#endif
+
+  texture_info.free();
+}
+
+bool CPUDevice::show_samples() const
+{
+  return (info.cpu_threads == 1);
+}
+
+BVHLayoutMask CPUDevice::get_bvh_layout_mask() const
+{
+  BVHLayoutMask bvh_layout_mask = BVH_LAYOUT_BVH2;
+#ifdef WITH_EMBREE
+  bvh_layout_mask |= BVH_LAYOUT_EMBREE;
+#endif /* WITH_EMBREE */
+  return bvh_layout_mask;
+}
+
+bool CPUDevice::load_texture_info()
+{
+  if (!need_texture_info) {
+    return false;
+  }
+
+  texture_info.copy_to_device();
+  need_texture_info = false;
+
+  return true;
+}
+
+void CPUDevice::mem_alloc(device_memory &mem)
+{
+  if (mem.type == MEM_TEXTURE) {
+    assert(!"mem_alloc not supported for textures.");
+  }
+  else if (mem.type == MEM_GLOBAL) {
+    assert(!"mem_alloc not supported for global memory.");
+  }
+  else {
+    if (mem.name) {
+      VLOG(1) << "Buffer allocate: " << mem.name << ", "
+              << string_human_readable_number(mem.memory_size()) << " bytes. ("
+              << string_human_readable_size(mem.memory_size()) << ")";
+    }
+
+    if (mem.type == MEM_DEVICE_ONLY) {
+      assert(!mem.host_pointer);
+      size_t alignment = MIN_ALIGNMENT_CPU_DATA_TYPES;
+      void *data = util_aligned_malloc(mem.memory_size(), alignment);
+      mem.device_pointer = (device_ptr)data;
+    }
+    else {
+      mem.device_pointer = (device_ptr)mem.host_pointer;
+    }
+
+    mem.device_size = mem.memory_size();
+    stats.mem_alloc(mem.device_size);
+  }
+}
+
+void CPUDevice::mem_copy_to(device_memory &mem)
+{
+  if (mem.type == MEM_GLOBAL) {
+    global_free(mem);
+    global_alloc(mem);
+  }
+  else if (mem.type == MEM_TEXTURE) {
+    tex_free((device_texture &)mem);
+    tex_alloc((device_texture &)mem);
+  }
+  else {
+    if (!mem.device_pointer) {
+      mem_alloc(mem);
+    }
+
+    /* copy is no-op */
+  }
+}
+
+void CPUDevice::mem_copy_from(
+    device_memory & /*mem*/, size_t /*y*/, size_t /*w*/, size_t /*h*/, size_t /*elem*/)
+{
+  /* no-op */
+}
+
+void CPUDevice::mem_zero(device_memory &mem)
+{
+  if (!mem.device_pointer) {
+    mem_alloc(mem);
+  }
+
+  if (mem.device_pointer) {
+    memset((void *)mem.device_pointer, 0, mem.memory_size());
+  }
+}
+
+void CPUDevice::mem_free(device_memory &mem)
+{
+  if (mem.type == MEM_GLOBAL) {
+    global_free(mem);
+  }
+  else if (mem.type == MEM_TEXTURE) {
+    tex_free((device_texture &)mem);
+  }
+  else if (mem.device_pointer) {
+    if (mem.type == MEM_DEVICE_ONLY) {
+      util_aligned_free((void *)mem.device_pointer);
+    }
+    mem.device_pointer = 0;
+    stats.mem_free(mem.device_size);
+    mem.device_size = 0;
+  }
+}
+
+device_ptr CPUDevice::mem_alloc_sub_ptr(device_memory &mem, size_t offset, size_t /*size*/)
+{
+  return (device_ptr)(((char *)mem.device_pointer) + mem.memory_elements_size(offset));
+}
+
+void CPUDevice::const_copy_to(const char *name, void *host, size_t size)
+{
+#if WITH_EMBREE
+  if (strcmp(name, "__data") == 0) {
+    assert(size <= sizeof(KernelData));
+
+    // Update scene handle (since it is different for each device on multi devices)
+    KernelData *const data = (KernelData *)host;
+    data->bvh.scene = embree_scene;
+  }
+#endif
+  kernel_const_copy(&kernel_globals, name, host, size);
+}
+
+void CPUDevice::global_alloc(device_memory &mem)
+{
+  VLOG(1) << "Global memory allocate: " << mem.name << ", "
+          << string_human_readable_number(mem.memory_size()) << " bytes. ("
+          << string_human_readable_size(mem.memory_size()) << ")";
+
+  kernel_global_memory_copy(&kernel_globals, mem.name, mem.host_pointer, mem.data_size);
+
+  mem.device_pointer = (device_ptr)mem.host_pointer;
+  mem.device_size = mem.memory_size();
+  stats.mem_alloc(mem.device_size);
+}
+
+void CPUDevice::global_free(device_memory &mem)
+{
+  if (mem.device_pointer) {
+    mem.device_pointer = 0;
+    stats.mem_free(mem.device_size);
+    mem.device_size = 0;
+  }
+}
+
+void CPUDevice::tex_alloc(device_texture &mem)
+{
+  VLOG(1) << "Texture allocate: " << mem.name << ", "
+          << string_human_readable_number(mem.memory_size()) << " bytes. ("
+          << string_human_readable_size(mem.memory_size()) << ")";
+
+  mem.device_pointer = (device_ptr)mem.host_pointer;
+  mem.device_size = mem.memory_size();
+  stats.mem_alloc(mem.device_size);
+
+  const uint slot = mem.slot;
+  if (slot >= texture_info.size()) {
+    /* Allocate some slots in advance, to reduce amount of re-allocations. */
+    texture_info.resize(slot + 128);
+  }
+
+  texture_info[slot] = mem.info;
+  texture_info[slot].data = (uint64_t)mem.host_pointer;
+  need_texture_info = true;
+}
+
+void CPUDevice::tex_free(device_texture &mem)
+{
+  if (mem.device_pointer) {
+    mem.device_pointer = 0;
+    stats.mem_free(mem.device_size);
+    mem.device_size = 0;
+    need_texture_info = true;
+  }
+}
+
+void CPUDevice::build_bvh(BVH *bvh, Progress &progress, bool refit)
+{
+#ifdef WITH_EMBREE
+  if (bvh->params.bvh_layout == BVH_LAYOUT_EMBREE ||
+      bvh->params.bvh_layout == BVH_LAYOUT_MULTI_OPTIX_EMBREE) {
+    BVHEmbree *const bvh_embree = static_cast<BVHEmbree *>(bvh);
+    if (refit) {
+      bvh_embree->refit(progress);
+    }
+    else {
+      bvh_embree->build(progress, &stats, embree_device);
+    }
+
+    if (bvh->params.top_level) {
+      embree_scene = bvh_embree->scene;
+    }
+  }
+  else
+#endif
+    Device::build_bvh(bvh, progress, refit);
+}
+
+const CPUKernels *CPUDevice::get_cpu_kernels() const
+{
+  return &kernels;
+}
+
+void CPUDevice::get_cpu_kernel_thread_globals(
+    vector<CPUKernelThreadGlobals> &kernel_thread_globals)
+{
+  /* Ensure latest texture info is loaded into kernel globals before returning. */
+  load_texture_info();
+
+  kernel_thread_globals.clear();
+  void *osl_memory = get_cpu_osl_memory();
+  for (int i = 0; i < info.cpu_threads; i++) {
+    kernel_thread_globals.emplace_back(kernel_globals, osl_memory, profiler);
+  }
+}
+
+void *CPUDevice::get_cpu_osl_memory()
+{
+#ifdef WITH_OSL
+  return &osl_globals;
+#else
+  return NULL;
+#endif
+}
+
+bool CPUDevice::load_kernels(const uint /*kernel_features*/)
+{
+  return true;
+}
+
+CCL_NAMESPACE_END

intern/cycles/device/cpu/device_impl.h

@@ -0,0 +1,102 @@
+/*
+ * Copyright 2011-2021 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
+
+/* So ImathMath is included before our kernel_cpu_compat. */
+#ifdef WITH_OSL
+/* So no context pollution happens from indirectly included windows.h */
+#  include "util/util_windows.h"
+#  include <OSL/oslexec.h>
+#endif
+
+#ifdef WITH_EMBREE
+#  include <embree3/rtcore.h>
+#endif
+
+#include "device/cpu/kernel.h"
+#include "device/device.h"
+#include "device/device_memory.h"
+
+// clang-format off
+#include "kernel/device/cpu/compat.h"
+#include "kernel/device/cpu/kernel.h"
+#include "kernel/device/cpu/globals.h"
+
+#include "kernel/osl/osl_shader.h"
+#include "kernel/osl/osl_globals.h"
+// clang-format on
+
+CCL_NAMESPACE_BEGIN
+
+class CPUDevice : public Device {
+ public:
+  KernelGlobals kernel_globals;
+
+  device_vector<TextureInfo> texture_info;
+  bool need_texture_info;
+
+#ifdef WITH_OSL
+  OSLGlobals osl_globals;
+#endif
+#ifdef WITH_EMBREE
+  RTCScene embree_scene = NULL;
+  RTCDevice embree_device;
+#endif
+
+  CPUKernels kernels;
+
+  CPUDevice(const DeviceInfo &info_, Stats &stats_, Profiler &profiler_);
+  ~CPUDevice();
+
+  virtual bool show_samples() const override;
+
+  virtual BVHLayoutMask get_bvh_layout_mask() const override;
+
+  /* Returns true if the texture info was copied to the device (meaning, some more
+   * re-initialization might be needed). */
+  bool load_texture_info();
+
+  virtual void mem_alloc(device_memory &mem) override;
+  virtual void mem_copy_to(device_memory &mem) override;
+  virtual void mem_copy_from(
+      device_memory &mem, size_t y, size_t w, size_t h, size_t elem) override;
+  virtual void mem_zero(device_memory &mem) override;
+  virtual void mem_free(device_memory &mem) override;
+  virtual device_ptr mem_alloc_sub_ptr(device_memory &mem,
+                                       size_t offset,
+                                       size_t /*size*/) override;
+
+  virtual void const_copy_to(const char *name, void *host, size_t size) override;
+
+  void global_alloc(device_memory &mem);
+  void global_free(device_memory &mem);
+
+  void tex_alloc(device_texture &mem);
+  void tex_free(device_texture &mem);
+
+  void build_bvh(BVH *bvh, Progress &progress, bool refit) override;
+
+  virtual const CPUKernels *get_cpu_kernels() const override;
+  virtual void get_cpu_kernel_thread_globals(
+      vector<CPUKernelThreadGlobals> &kernel_thread_globals) override;
+  virtual void *get_cpu_osl_memory() override;
+
+ protected:
+  virtual bool load_kernels(uint /*kernel_features*/) override;
+};
+
+CCL_NAMESPACE_END

intern/cycles/device/cpu/kernel.cpp

@@ -0,0 +1,61 @@
+/*
+ * Copyright 2011-2021 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.
+ */
+
+#include "device/cpu/kernel.h"
+
+#include "kernel/device/cpu/kernel.h"
+
+CCL_NAMESPACE_BEGIN
+
+#define KERNEL_FUNCTIONS(name) \
+  KERNEL_NAME_EVAL(cpu, name), KERNEL_NAME_EVAL(cpu_sse2, name), \
+      KERNEL_NAME_EVAL(cpu_sse3, name), KERNEL_NAME_EVAL(cpu_sse41, name), \
+      KERNEL_NAME_EVAL(cpu_avx, name), KERNEL_NAME_EVAL(cpu_avx2, name)
+
+#define REGISTER_KERNEL(name) name(KERNEL_FUNCTIONS(name))
+
+CPUKernels::CPUKernels()
+    : /* Integrator. */
+      REGISTER_KERNEL(integrator_init_from_camera),
+      REGISTER_KERNEL(integrator_init_from_bake),
+      REGISTER_KERNEL(integrator_intersect_closest),
+      REGISTER_KERNEL(integrator_intersect_shadow),
+      REGISTER_KERNEL(integrator_intersect_subsurface),
+      REGISTER_KERNEL(integrator_intersect_volume_stack),
+      REGISTER_KERNEL(integrator_shade_background),
+      REGISTER_KERNEL(integrator_shade_light),
+      REGISTER_KERNEL(integrator_shade_shadow),
+      REGISTER_KERNEL(integrator_shade_surface),
+      REGISTER_KERNEL(integrator_shade_volume),
+      REGISTER_KERNEL(integrator_megakernel),
+      /* Shader evaluation. */
+      REGISTER_KERNEL(shader_eval_displace),
+      REGISTER_KERNEL(shader_eval_background),
+      /* Adaptive sampling. */
+      REGISTER_KERNEL(adaptive_sampling_convergence_check),
+      REGISTER_KERNEL(adaptive_sampling_filter_x),
+      REGISTER_KERNEL(adaptive_sampling_filter_y),
+      /* Cryptomatte. */
+      REGISTER_KERNEL(cryptomatte_postprocess),
+      /* Bake. */
+      REGISTER_KERNEL(bake)
+{
+}
+
+#undef REGISTER_KERNEL
+#undef KERNEL_FUNCTIONS
+
+CCL_NAMESPACE_END

intern/cycles/device/cpu/kernel.h

@@ -0,0 +1,111 @@
+/*
+ * Copyright 2011-2021 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
+
+#include "device/cpu/kernel_function.h"
+#include "util/util_types.h"
+
+CCL_NAMESPACE_BEGIN
+
+struct KernelGlobals;
+struct IntegratorStateCPU;
+struct TileInfo;
+
+class CPUKernels {
+ public:
+  /* Integrator. */
+
+  using IntegratorFunction =
+      CPUKernelFunction<void (*)(const KernelGlobals *kg, IntegratorStateCPU *state)>;
+  using IntegratorShadeFunction = CPUKernelFunction<void (*)(
+      const KernelGlobals *kg, IntegratorStateCPU *state, ccl_global float *render_buffer)>;
+  using IntegratorInitFunction = CPUKernelFunction<bool (*)(const KernelGlobals *kg,
+                                                            IntegratorStateCPU *state,
+                                                            KernelWorkTile *tile,
+                                                            ccl_global float *render_buffer)>;
+
+  IntegratorInitFunction integrator_init_from_camera;
+  IntegratorInitFunction integrator_init_from_bake;
+  IntegratorFunction integrator_intersect_closest;
+  IntegratorFunction integrator_intersect_shadow;
+  IntegratorFunction integrator_intersect_subsurface;
+  IntegratorFunction integrator_intersect_volume_stack;
+  IntegratorShadeFunction integrator_shade_background;
+  IntegratorShadeFunction integrator_shade_light;
+  IntegratorShadeFunction integrator_shade_shadow;
+  IntegratorShadeFunction integrator_shade_surface;
+  IntegratorShadeFunction integrator_shade_volume;
+  IntegratorShadeFunction integrator_megakernel;
+
+  /* Shader evaluation. */
+
+  using ShaderEvalFunction = CPUKernelFunction<void (*)(
+      const KernelGlobals *kg, const KernelShaderEvalInput *, float4 *, const int)>;
+
+  ShaderEvalFunction shader_eval_displace;
+  ShaderEvalFunction shader_eval_background;
+
+  /* Adaptive stopping. */
+
+  using AdaptiveSamplingConvergenceCheckFunction =
+      CPUKernelFunction<bool (*)(const KernelGlobals *kg,
+                                 ccl_global float *render_buffer,
+                                 int x,
+                                 int y,
+                                 float threshold,
+                                 bool reset,
+                                 int offset,
+                                 int stride)>;
+
+  using AdaptiveSamplingFilterXFunction =
+      CPUKernelFunction<void (*)(const KernelGlobals *kg,
+                                 ccl_global float *render_buffer,
+                                 int y,
+                                 int start_x,
+                                 int width,
+                                 int offset,
+                                 int stride)>;
+
+  using AdaptiveSamplingFilterYFunction =
+      CPUKernelFunction<void (*)(const KernelGlobals *kg,
+                                 ccl_global float *render_buffer,
+                                 int x,
+                                 int start_y,
+                                 int height,
+                                 int offset,
+                                 int stride)>;
+
+  AdaptiveSamplingConvergenceCheckFunction adaptive_sampling_convergence_check;
+
+  AdaptiveSamplingFilterXFunction adaptive_sampling_filter_x;
+  AdaptiveSamplingFilterYFunction adaptive_sampling_filter_y;
+
+  /* Cryptomatte. */
+
+  using CryptomattePostprocessFunction = CPUKernelFunction<void (*)(
+      const KernelGlobals *kg, ccl_global float *render_buffer, int pixel_index)>;
+
+  CryptomattePostprocessFunction cryptomatte_postprocess;
+
+  /* Bake. */
+
+  CPUKernelFunction<void (*)(const KernelGlobals *, float *, int, int, int, int, int)> bake;
+
+  CPUKernels();
+};
+
+CCL_NAMESPACE_END

intern/cycles/device/cpu/kernel_function.h

@@ -0,0 +1,124 @@
+/*
+ * Copyright 2011-2021 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
+
+#include "util/util_debug.h"
+#include "util/util_system.h"
+
+CCL_NAMESPACE_BEGIN
+
+/* A wrapper around per-microarchitecture variant of a kernel function.
+ *
+ * Provides a function-call-like API which gets routed to the most suitable implementation.
+ *
+ * For example, on a computer which only has SSE4.1 the kernel_sse41 will be used. */
+template<typename FunctionType> class CPUKernelFunction {
+ public:
+  CPUKernelFunction(FunctionType kernel_default,
+                    FunctionType kernel_sse2,
+                    FunctionType kernel_sse3,
+                    FunctionType kernel_sse41,
+                    FunctionType kernel_avx,
+                    FunctionType kernel_avx2)
+  {
+    kernel_info_ = get_best_kernel_info(
+        kernel_default, kernel_sse2, kernel_sse3, kernel_sse41, kernel_avx, kernel_avx2);
+  }
+
+  template<typename... Args> inline auto operator()(Args... args) const
+  {
+    assert(kernel_info_.kernel);
+
+    return kernel_info_.kernel(args...);
+  }
+
+  const char *get_uarch_name() const
+  {
+    return kernel_info_.uarch_name;
+  }
+
+ protected:
+  /* Helper class which allows to pass human-readable microarchitecture name together with function
+   * pointer. */
+  class KernelInfo {
+   public:
+    KernelInfo() : KernelInfo("", nullptr)
+    {
+    }
+
+    /* TODO(sergey): Use string view, to have higher-level functionality (i.e. comparison) without
+     * memory allocation. */
+    KernelInfo(const char *uarch_name, FunctionType kernel)
+        : uarch_name(uarch_name), kernel(kernel)
+    {
+    }
+
+    const char *uarch_name;
+    FunctionType kernel;
+  };
+
+  KernelInfo get_best_kernel_info(FunctionType kernel_default,
+                                  FunctionType kernel_sse2,
+                                  FunctionType kernel_sse3,
+                                  FunctionType kernel_sse41,
+                                  FunctionType kernel_avx,
+                                  FunctionType kernel_avx2)
+  {
+    /* Silence warnings about unused variables when compiling without some architectures. */
+    (void)kernel_sse2;
+    (void)kernel_sse3;
+    (void)kernel_sse41;
+    (void)kernel_avx;
+    (void)kernel_avx2;
+
+#ifdef WITH_CYCLES_OPTIMIZED_KERNEL_AVX2
+    if (DebugFlags().cpu.has_avx2() && system_cpu_support_avx2()) {
+      return KernelInfo("AVX2", kernel_avx2);
+    }
+#endif
+
+#ifdef WITH_CYCLES_OPTIMIZED_KERNEL_AVX
+    if (DebugFlags().cpu.has_avx() && system_cpu_support_avx()) {
+      return KernelInfo("AVX", kernel_avx);
+    }
+#endif
+
+#ifdef WITH_CYCLES_OPTIMIZED_KERNEL_SSE41
+    if (DebugFlags().cpu.has_sse41() && system_cpu_support_sse41()) {
+      return KernelInfo("SSE4.1", kernel_sse41);
+    }
+#endif
+
+#ifdef WITH_CYCLES_OPTIMIZED_KERNEL_SSE3
+    if (DebugFlags().cpu.has_sse3() && system_cpu_support_sse3()) {
+      return KernelInfo("SSE3", kernel_sse3);
+    }
+#endif
+
+#ifdef WITH_CYCLES_OPTIMIZED_KERNEL_SSE2
+    if (DebugFlags().cpu.has_sse2() && system_cpu_support_sse2()) {
+      return KernelInfo("SSE2", kernel_sse2);
+    }
+#endif
+
+    return KernelInfo("default", kernel_default);
+  }
+
+  KernelInfo kernel_info_;
+};
+
+CCL_NAMESPACE_END

intern/cycles/device/cpu/kernel_thread_globals.cpp

@@ -0,0 +1,85 @@
+/*
+ * Copyright 2011-2021 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.
+ */
+
+#include "device/cpu/kernel_thread_globals.h"
+
+// clang-format off
+#include "kernel/osl/osl_shader.h"
+#include "kernel/osl/osl_globals.h"
+// clang-format on
+
+#include "util/util_profiling.h"
+
+CCL_NAMESPACE_BEGIN
+
+CPUKernelThreadGlobals::CPUKernelThreadGlobals(const KernelGlobals &kernel_globals,
+                                               void *osl_globals_memory,
+                                               Profiler &cpu_profiler)
+    : KernelGlobals(kernel_globals), cpu_profiler_(cpu_profiler)
+{
+  reset_runtime_memory();
+
+#ifdef WITH_OSL
+  OSLShader::thread_init(this, reinterpret_cast<OSLGlobals *>(osl_globals_memory));
+#else
+  (void)osl_globals_memory;
+#endif
+}
+
+CPUKernelThreadGlobals::CPUKernelThreadGlobals(CPUKernelThreadGlobals &&other) noexcept
+    : KernelGlobals(std::move(other)), cpu_profiler_(other.cpu_profiler_)
+{
+  other.reset_runtime_memory();
+}
+
+CPUKernelThreadGlobals::~CPUKernelThreadGlobals()
+{
+#ifdef WITH_OSL
+  OSLShader::thread_free(this);
+#endif
+}
+
+CPUKernelThreadGlobals &CPUKernelThreadGlobals::operator=(CPUKernelThreadGlobals &&other)
+{
+  if (this == &other) {
+    return *this;
+  }
+
+  *static_cast<KernelGlobals *>(this) = *static_cast<KernelGlobals *>(&other);
+
+  other.reset_runtime_memory();
+
+  return *this;
+}
+
+void CPUKernelThreadGlobals::reset_runtime_memory()
+{
+#ifdef WITH_OSL
+  osl = nullptr;
+#endif
+}
+
+void CPUKernelThreadGlobals::start_profiling()
+{
+  cpu_profiler_.add_state(&profiler);
+}
+
+void CPUKernelThreadGlobals::stop_profiling()
+{
+  cpu_profiler_.remove_state(&profiler);
+}
+
+CCL_NAMESPACE_END

intern/cycles/device/cpu/kernel_thread_globals.h

@@ -0,0 +1,57 @@
+/*
+ * Copyright 2011-2021 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
+
+#include "kernel/device/cpu/compat.h"
+#include "kernel/device/cpu/globals.h"
+
+CCL_NAMESPACE_BEGIN
+
+class Profiler;
+
+/* A special class which extends memory ownership of the `KernelGlobals` decoupling any resource
+ * which is not thread-safe for access. Every worker thread which needs to operate on
+ * `KernelGlobals` needs to initialize its own copy of this object.
+ *
+ * NOTE: Only minimal subset of objects are copied: `KernelData` is never copied. This means that
+ * there is no unnecessary data duplication happening when using this object. */
+class CPUKernelThreadGlobals : public KernelGlobals {
+ public:
+  /* TODO(sergey): Would be nice to have properly typed OSLGlobals even in the case when building
+   * without OSL support. Will avoid need to those unnamed pointers and casts. */
+  CPUKernelThreadGlobals(const KernelGlobals &kernel_globals,
+                         void *osl_globals_memory,
+                         Profiler &cpu_profiler);
+
+  ~CPUKernelThreadGlobals();
+
+  CPUKernelThreadGlobals(const CPUKernelThreadGlobals &other) = delete;
+  CPUKernelThreadGlobals(CPUKernelThreadGlobals &&other) noexcept;
+
+  CPUKernelThreadGlobals &operator=(const CPUKernelThreadGlobals &other) = delete;
+  CPUKernelThreadGlobals &operator=(CPUKernelThreadGlobals &&other);
+
+  void start_profiling();
+  void stop_profiling();
+
+ protected:
+  void reset_runtime_memory();
+
+  Profiler &cpu_profiler_;
+};
+
+CCL_NAMESPACE_END

intern/cycles/device/cuda/device.cpp

@@ -14,21 +14,25 @@
  * limitations under the License.
  */
 
+#include "device/cuda/device.h"
+
+#include "util/util_logging.h"
+
 #ifdef WITH_CUDA
-
-#  include "device/cuda/device_cuda.h"
+#  include "device/cuda/device_impl.h"
 #  include "device/device.h"
-#  include "device/device_intern.h"
 
-#  include "util/util_logging.h"
 #  include "util/util_string.h"
 #  include "util/util_windows.h"
+#endif /* WITH_CUDA */
 
 CCL_NAMESPACE_BEGIN
 
 bool device_cuda_init()
 {
-#  ifdef WITH_CUDA_DYNLOAD
+#if !defined(WITH_CUDA)
+  return false;
+#elif defined(WITH_CUDA_DYNLOAD)
   static bool initialized = false;
   static bool result = false;
 
@@ -59,16 +63,27 @@ bool device_cuda_init()
   }
 
   return result;
-#  else  /* WITH_CUDA_DYNLOAD */
+#else  /* WITH_CUDA_DYNLOAD */
   return true;
-#  endif /* WITH_CUDA_DYNLOAD */
+#endif /* WITH_CUDA_DYNLOAD */
 }
 
-Device *device_cuda_create(DeviceInfo &info, Stats &stats, Profiler &profiler, bool background)
+Device *device_cuda_create(const DeviceInfo &info, Stats &stats, Profiler &profiler)
 {
-  return new CUDADevice(info, stats, profiler, background);
+#ifdef WITH_CUDA
+  return new CUDADevice(info, stats, profiler);
+#else
+  (void)info;
+  (void)stats;
+  (void)profiler;
+
+  LOG(FATAL) << "Request to create CUDA device without compiled-in support. Should never happen.";
+
+  return nullptr;
+#endif
 }
 
+#ifdef WITH_CUDA
 static CUresult device_cuda_safe_init()
 {
 #  ifdef _WIN32
@@ -86,9 +101,11 @@ static CUresult device_cuda_safe_init()
   return cuInit(0);
 #  endif
 }
+#endif /* WITH_CUDA */
 
 void device_cuda_info(vector<DeviceInfo> &devices)
 {
+#ifdef WITH_CUDA
   CUresult result = device_cuda_safe_init();
   if (result != CUDA_SUCCESS) {
     if (result != CUDA_ERROR_NO_DEVICE)
@@ -129,9 +146,9 @@ void device_cuda_info(vector<DeviceInfo> &devices)
 
     info.has_half_images = (major >= 3);
     info.has_nanovdb = true;
-    info.has_volume_decoupled = false;
-    info.has_adaptive_stop_per_sample = false;
-    info.denoisers = DENOISER_NLM;
+    info.denoisers = 0;
+
+    info.has_gpu_queue = true;
 
     /* Check if the device has P2P access to any other device in the system. */
     for (int peer_num = 0; peer_num < count && !info.has_peer_memory; peer_num++) {
@@ -182,10 +199,14 @@ void device_cuda_info(vector<DeviceInfo> &devices)
 
   if (!display_devices.empty())
     devices.insert(devices.end(), display_devices.begin(), display_devices.end());
+#else  /* WITH_CUDA */
+  (void)devices;
+#endif /* WITH_CUDA */
 }
 
 string device_cuda_capabilities()
 {
+#ifdef WITH_CUDA
   CUresult result = device_cuda_safe_init();
   if (result != CUDA_SUCCESS) {
     if (result != CUDA_ERROR_NO_DEVICE) {
@@ -310,8 +331,10 @@ string device_cuda_capabilities()
   }
 
   return capabilities;
+
+#else  /* WITH_CUDA */
+  return "";
+#endif /* WITH_CUDA */
 }
 
 CCL_NAMESPACE_END
-
-#endif

intern/cycles/device/cuda/device.h

@@ -0,0 +1,37 @@
+/*
+ * Copyright 2011-2021 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
+
+#include "util/util_string.h"
+#include "util/util_vector.h"
+
+CCL_NAMESPACE_BEGIN
+
+class Device;
+class DeviceInfo;
+class Profiler;
+class Stats;
+
+bool device_cuda_init();
+
+Device *device_cuda_create(const DeviceInfo &info, Stats &stats, Profiler &profiler);
+
+void device_cuda_info(vector<DeviceInfo> &devices);
+
+string device_cuda_capabilities();
+
+CCL_NAMESPACE_END

intern/cycles/device/cuda/device_cuda.h

@@ -1,270 +0,0 @@
-/*
- * Copyright 2011-2013 Blender Foundation
- *
- * Licensed under the Apache License, Version 2.0 (the "License");
- * you may not use this file except in compliance with the License.
- * You may obtain a copy of the License at
- *
- * http://www.apache.org/licenses/LICENSE-2.0
- *
- * Unless required by applicable law or agreed to in writing, software
- * distributed under the License is distributed on an "AS IS" BASIS,
- * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
- * See the License for the specific language governing permissions and
- * limitations under the License.
- */
-
-#ifdef WITH_CUDA
-
-#  include "device/device.h"
-#  include "device/device_denoising.h"
-#  include "device/device_split_kernel.h"
-
-#  include "util/util_map.h"
-#  include "util/util_task.h"
-
-#  ifdef WITH_CUDA_DYNLOAD
-#    include "cuew.h"
-#  else
-#    include "util/util_opengl.h"
-#    include <cuda.h>
-#    include <cudaGL.h>
-#  endif
-
-CCL_NAMESPACE_BEGIN
-
-class CUDASplitKernel;
-
-class CUDADevice : public Device {
-
-  friend class CUDASplitKernelFunction;
-  friend class CUDASplitKernel;
-  friend class CUDAContextScope;
-
- public:
-  DedicatedTaskPool task_pool;
-  CUdevice cuDevice;
-  CUcontext cuContext;
-  CUmodule cuModule, cuFilterModule;
-  size_t device_texture_headroom;
-  size_t device_working_headroom;
-  bool move_texture_to_host;
-  size_t map_host_used;
-  size_t map_host_limit;
-  int can_map_host;
-  int pitch_alignment;
-  int cuDevId;
-  int cuDevArchitecture;
-  bool first_error;
-  CUDASplitKernel *split_kernel;
-
-  struct CUDAMem {
-    CUDAMem() : texobject(0), array(0), use_mapped_host(false)
-    {
-    }
-
-    CUtexObject texobject;
-    CUarray array;
-
-    /* If true, a mapped host memory in shared_pointer is being used. */
-    bool use_mapped_host;
-  };
-  typedef map<device_memory *, CUDAMem> CUDAMemMap;
-  CUDAMemMap cuda_mem_map;
-  thread_mutex cuda_mem_map_mutex;
-
-  struct PixelMem {
-    GLuint cuPBO;
-    CUgraphicsResource cuPBOresource;
-    GLuint cuTexId;
-    int w, h;
-  };
-  map<device_ptr, PixelMem> pixel_mem_map;
-
-  /* Bindless Textures */
-  device_vector<TextureInfo> texture_info;
-  bool need_texture_info;
-
-  /* Kernels */
-  struct {
-    bool loaded;
-
-    CUfunction adaptive_stopping;
-    CUfunction adaptive_filter_x;
-    CUfunction adaptive_filter_y;
-    CUfunction adaptive_scale_samples;
-    int adaptive_num_threads_per_block;
-  } functions;
-
-  static bool have_precompiled_kernels();
-
-  virtual bool show_samples() const override;
-
-  virtual BVHLayoutMask get_bvh_layout_mask() const override;
-
-  void set_error(const string &error) override;
-
-  CUDADevice(DeviceInfo &info, Stats &stats, Profiler &profiler, bool background_);
-
-  virtual ~CUDADevice();
-
-  bool support_device(const DeviceRequestedFeatures & /*requested_features*/);
-
-  bool check_peer_access(Device *peer_device) override;
-
-  bool use_adaptive_compilation();
-
-  bool use_split_kernel();
-
-  virtual string compile_kernel_get_common_cflags(
-      const DeviceRequestedFeatures &requested_features, bool filter = false, bool split = false);
-
-  string compile_kernel(const DeviceRequestedFeatures &requested_features,
-                        const char *name,
-                        const char *base = "cuda",
-                        bool force_ptx = false);
-
-  virtual bool load_kernels(const DeviceRequestedFeatures &requested_features) override;
-
-  void load_functions();
-
-  void reserve_local_memory(const DeviceRequestedFeatures &requested_features);
-
-  void init_host_memory();
-
-  void load_texture_info();
-
-  void move_textures_to_host(size_t size, bool for_texture);
-
-  CUDAMem *generic_alloc(device_memory &mem, size_t pitch_padding = 0);
-
-  void generic_copy_to(device_memory &mem);
-
-  void generic_free(device_memory &mem);
-
-  void mem_alloc(device_memory &mem) override;
-
-  void mem_copy_to(device_memory &mem) override;
-
-  void mem_copy_from(device_memory &mem, int y, int w, int h, int elem) override;
-
-  void mem_zero(device_memory &mem) override;
-
-  void mem_free(device_memory &mem) override;
-
-  device_ptr mem_alloc_sub_ptr(device_memory &mem, int offset, int /*size*/) override;
-
-  virtual void const_copy_to(const char *name, void *host, size_t size) override;
-
-  void global_alloc(device_memory &mem);
-
-  void global_free(device_memory &mem);
-
-  void tex_alloc(device_texture &mem);
-
-  void tex_free(device_texture &mem);
-
-  bool denoising_non_local_means(device_ptr image_ptr,
-                                 device_ptr guide_ptr,
-                                 device_ptr variance_ptr,
-                                 device_ptr out_ptr,
-                                 DenoisingTask *task);
-
-  bool denoising_construct_transform(DenoisingTask *task);
-
-  bool denoising_accumulate(device_ptr color_ptr,
-                            device_ptr color_variance_ptr,
-                            device_ptr scale_ptr,
-                            int frame,
-                            DenoisingTask *task);
-
-  bool denoising_solve(device_ptr output_ptr, DenoisingTask *task);
-
-  bool denoising_combine_halves(device_ptr a_ptr,
-                                device_ptr b_ptr,
-                                device_ptr mean_ptr,
-                                device_ptr variance_ptr,
-                                int r,
-                                int4 rect,
-                                DenoisingTask *task);
-
-  bool denoising_divide_shadow(device_ptr a_ptr,
-                               device_ptr b_ptr,
-                               device_ptr sample_variance_ptr,
-                               device_ptr sv_variance_ptr,
-                               device_ptr buffer_variance_ptr,
-                               DenoisingTask *task);
-
-  bool denoising_get_feature(int mean_offset,
-                             int variance_offset,
-                             device_ptr mean_ptr,
-                             device_ptr variance_ptr,
-                             float scale,
-                             DenoisingTask *task);
-
-  bool denoising_write_feature(int out_offset,
-                               device_ptr from_ptr,
-                               device_ptr buffer_ptr,
-                               DenoisingTask *task);
-
-  bool denoising_detect_outliers(device_ptr image_ptr,
-                                 device_ptr variance_ptr,
-                                 device_ptr depth_ptr,
-                                 device_ptr output_ptr,
-                                 DenoisingTask *task);
-
-  void denoise(RenderTile &rtile, DenoisingTask &denoising);
-
-  void adaptive_sampling_filter(uint filter_sample,
-                                WorkTile *wtile,
-                                CUdeviceptr d_wtile,
-                                CUstream stream = 0);
-  void adaptive_sampling_post(RenderTile &rtile,
-                              WorkTile *wtile,
-                              CUdeviceptr d_wtile,
-                              CUstream stream = 0);
-
-  void render(DeviceTask &task, RenderTile &rtile, device_vector<WorkTile> &work_tiles);
-
-  void film_convert(DeviceTask &task,
-                    device_ptr buffer,
-                    device_ptr rgba_byte,
-                    device_ptr rgba_half);
-
-  void shader(DeviceTask &task);
-
-  CUdeviceptr map_pixels(device_ptr mem);
-
-  void unmap_pixels(device_ptr mem);
-
-  void pixels_alloc(device_memory &mem);
-
-  void pixels_copy_from(device_memory &mem, int y, int w, int h);
-
-  void pixels_free(device_memory &mem);
-
-  void draw_pixels(device_memory &mem,
-                   int y,
-                   int w,
-                   int h,
-                   int width,
-                   int height,
-                   int dx,
-                   int dy,
-                   int dw,
-                   int dh,
-                   bool transparent,
-                   const DeviceDrawParams &draw_params) override;
-
-  void thread_run(DeviceTask &task);
-
-  virtual void task_add(DeviceTask &task) override;
-
-  virtual void task_wait() override;
-
-  virtual void task_cancel() override;
-};
-
-CCL_NAMESPACE_END
-
-#endif

intern/cycles/device/cuda/device_impl.h

@@ -0,0 +1,154 @@
+/*
+ * Copyright 2011-2013 Blender Foundation
+ *
+ * Licensed under the Apache License, Version 2.0 (the "License");
+ * you may not use this file except in compliance with the License.
+ * You may obtain a copy of the License at
+ *
+ * http://www.apache.org/licenses/LICENSE-2.0
+ *
+ * Unless required by applicable law or agreed to in writing, software
+ * distributed under the License is distributed on an "AS IS" BASIS,
+ * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+ * See the License for the specific language governing permissions and
+ * limitations under the License.
+ */
+
+#ifdef WITH_CUDA
+
+#  include "device/cuda/kernel.h"
+#  include "device/cuda/queue.h"
+#  include "device/cuda/util.h"
+#  include "device/device.h"
+
+#  include "util/util_map.h"
+
+#  ifdef WITH_CUDA_DYNLOAD
+#    include "cuew.h"
+#  else
+#    include <cuda.h>
+#    include <cudaGL.h>
+#  endif
+
+CCL_NAMESPACE_BEGIN
+
+class DeviceQueue;
+
+class CUDADevice : public Device {
+
+  friend class CUDAContextScope;
+
+ public:
+  CUdevice cuDevice;
+  CUcontext cuContext;
+  CUmodule cuModule;
+  size_t device_texture_headroom;
+  size_t device_working_headroom;
+  bool move_texture_to_host;
+  size_t map_host_used;
+  size_t map_host_limit;
+  int can_map_host;
+  int pitch_alignment;
+  int cuDevId;
+  int cuDevArchitecture;
+  bool first_error;
+
+  struct CUDAMem {
+    CUDAMem() : texobject(0), array(0), use_mapped_host(false)
+    {
+    }
+
+    CUtexObject texobject;
+    CUarray array;
+
+    /* If true, a mapped host memory in shared_pointer is being used. */
+    bool use_mapped_host;
+  };
+  typedef map<device_memory *, CUDAMem> CUDAMemMap;
+  CUDAMemMap cuda_mem_map;
+  thread_mutex cuda_mem_map_mutex;
+
+  /* Bindless Textures */
+  device_vector<TextureInfo> texture_info;
+  bool need_texture_info;
+
+  CUDADeviceKernels kernels;
+
+  static bool have_precompiled_kernels();
+
+  virtual bool show_samples() const override;
+
+  virtual BVHLayoutMask get_bvh_layout_mask() const override;
+
+  void set_error(const string &error) override;
+
+  CUDADevice(const DeviceInfo &info, Stats &stats, Profiler &profiler);
+
+  virtual ~CUDADevice();
+
+  bool support_device(const uint /*kernel_features*/);
+
+  bool check_peer_access(Device *peer_device) override;
+
+  bool use_adaptive_compilation();
+
+  virtual string compile_kernel_get_common_cflags(const uint kernel_features);
+
+  string compile_kernel(const uint kernel_features,
+                        const char *name,
+                        const char *base = "cuda",
+                        bool force_ptx = false);
+
+  virtual bool load_kernels(const uint kernel_features) override;
+
+  void reserve_local_memory(const uint kernel_features);
+
+  void init_host_memory();
+
+  void load_texture_info();
+
+  void move_textures_to_host(size_t size, bool for_texture);
+
+  CUDAMem *generic_alloc(device_memory &mem, size_t pitch_padding = 0);
+
+  void generic_copy_to(device_memory &mem);
+
+  void generic_free(device_memory &mem);
+
+  void mem_alloc(device_memory &mem) override;
+
+  void mem_copy_to(device_memory &mem) override;
+
+  void mem_copy_from(device_memory &mem, size_t y, size_t w, size_t h, size_t elem) override;
+
+  void mem_zero(device_memory &mem) override;
+
+  void mem_free(device_memory &mem) override;
+
+  device_ptr mem_alloc_sub_ptr(device_memory &mem, size_t offset, size_t /*size*/) override;
+
+  virtual void const_copy_to(const char *name, void *host, size_t size) override;
+
+  void global_alloc(device_memory &mem);
+
+  void global_free(device_memory &mem);
+
+  void tex_alloc(device_texture &mem);
+
+  void tex_free(device_texture &mem);
+
+  virtual bool should_use_graphics_interop() override;
+
+  virtual unique_ptr<DeviceQueue> gpu_queue_create() override;
+
+  int get_num_multiprocessors();
+  int get_max_num_threads_per_multiprocessor();
+
+ protected:
+  bool get_device_attribute(CUdevice_attribute attribute, int *value);
+  int get_device_default_attribute(CUdevice_attribute attribute, int default_value);
+};
+
+CCL_NAMESPACE_END
+
+#endif

intern/cycles/device/cuda/graphics_interop.cpp

@@ -0,0 +1,102 @@
+/*
+ * Copyright 2011-2021 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.
+ */
+
+#ifdef WITH_CUDA
+
+#  include "device/cuda/graphics_interop.h"
+
+#  include "device/cuda/device_impl.h"
+#  include "device/cuda/util.h"
+
+CCL_NAMESPACE_BEGIN
+
+CUDADeviceGraphicsInterop::CUDADeviceGraphicsInterop(CUDADeviceQueue *queue)
+    : queue_(queue), device_(static_cast<CUDADevice *>(queue->device))
+{
+}
+
+CUDADeviceGraphicsInterop::~CUDADeviceGraphicsInterop()
+{
+  CUDAContextScope scope(device_);
+
+  if (cu_graphics_resource_) {
+    cuda_device_assert(device_, cuGraphicsUnregisterResource(cu_graphics_resource_));
+  }
+}
+
+void CUDADeviceGraphicsInterop::set_destination(
+    const DeviceGraphicsInteropDestination &destination)
+{
+  const int64_t new_buffer_area = int64_t(destination.buffer_width) * destination.buffer_height;
+
+  need_clear_ = destination.need_clear;
+
+  if (opengl_pbo_id_ == destination.opengl_pbo_id && buffer_area_ == new_buffer_area) {
+    return;
+  }
+
+  CUDAContextScope scope(device_);
+
+  if (cu_graphics_resource_) {
+    cuda_device_assert(device_, cuGraphicsUnregisterResource(cu_graphics_resource_));
+  }
+
+  const CUresult result = cuGraphicsGLRegisterBuffer(
+      &cu_graphics_resource_, destination.opengl_pbo_id, CU_GRAPHICS_MAP_RESOURCE_FLAGS_NONE);
+  if (result != CUDA_SUCCESS) {
+    LOG(ERROR) << "Error registering OpenGL buffer: " << cuewErrorString(result);
+  }
+
+  opengl_pbo_id_ = destination.opengl_pbo_id;
+  buffer_area_ = new_buffer_area;
+}
+
+device_ptr CUDADeviceGraphicsInterop::map()
+{
+  if (!cu_graphics_resource_) {
+    return 0;
+  }
+
+  CUDAContextScope scope(device_);
+
+  CUdeviceptr cu_buffer;
+  size_t bytes;
+
+  cuda_device_assert(device_, cuGraphicsMapResources(1, &cu_graphics_resource_, queue_->stream()));
+  cuda_device_assert(
+      device_, cuGraphicsResourceGetMappedPointer(&cu_buffer, &bytes, cu_graphics_resource_));
+
+  if (need_clear_) {
+    cuda_device_assert(
+        device_, cuMemsetD8Async(static_cast<CUdeviceptr>(cu_buffer), 0, bytes, queue_->stream()));
+
+    need_clear_ = false;
+  }
+
+  return static_cast<device_ptr>(cu_buffer);
+}
+
+void CUDADeviceGraphicsInterop::unmap()
+{
+  CUDAContextScope scope(device_);
+
+  cuda_device_assert(device_,
+                     cuGraphicsUnmapResources(1, &cu_graphics_resource_, queue_->stream()));
+}
+
+CCL_NAMESPACE_END
+
+#endif

intern/cycles/device/cuda/graphics_interop.h

@@ -0,0 +1,66 @@
+/*
+ * Copyright 2011-2021 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.
+ */
+
+#ifdef WITH_CUDA
+
+#  include "device/device_graphics_interop.h"
+
+#  ifdef WITH_CUDA_DYNLOAD
+#    include "cuew.h"
+#  else
+#    include <cuda.h>
+#  endif
+
+CCL_NAMESPACE_BEGIN
+
+class CUDADevice;
+class CUDADeviceQueue;
+
+class CUDADeviceGraphicsInterop : public DeviceGraphicsInterop {
+ public:
+  explicit CUDADeviceGraphicsInterop(CUDADeviceQueue *queue);
+
+  CUDADeviceGraphicsInterop(const CUDADeviceGraphicsInterop &other) = delete;
+  CUDADeviceGraphicsInterop(CUDADeviceGraphicsInterop &&other) noexcept = delete;
+
+  ~CUDADeviceGraphicsInterop();
+
+  CUDADeviceGraphicsInterop &operator=(const CUDADeviceGraphicsInterop &other) = delete;
+  CUDADeviceGraphicsInterop &operator=(CUDADeviceGraphicsInterop &&other) = delete;
+
+  virtual void set_destination(const DeviceGraphicsInteropDestination &destination) override;
+
+  virtual device_ptr map() override;
+  virtual void unmap() override;
+
+ protected:
+  CUDADeviceQueue *queue_ = nullptr;
+  CUDADevice *device_ = nullptr;
+
+  /* OpenGL PBO which is currently registered as the destination for the CUDA buffer. */
+  uint opengl_pbo_id_ = 0;
+  /* Buffer area in pixels of the corresponding PBO. */
+  int64_t buffer_area_ = 0;
+
+  /* The destination was requested to be cleared. */
+  bool need_clear_ = false;
+
+  CUgraphicsResource cu_graphics_resource_ = nullptr;
+};
+
+CCL_NAMESPACE_END
+
+#endif

intern/cycles/device/cuda/kernel.cpp

@@ -0,0 +1,69 @@
+/*
+ * Copyright 2011-2013 Blender Foundation
+ *
+ * Licensed under the Apache License, Version 2.0 (the "License");
+ * you may not use this file except in compliance with the License.
+ * You may obtain a copy of the License at
+ *
+ * http://www.apache.org/licenses/LICENSE-2.0
+ *
+ * Unless required by applicable law or agreed to in writing, software
+ * distributed under the License is distributed on an "AS IS" BASIS,
+ * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+ * See the License for the specific language governing permissions and
+ * limitations under the License.
+ */
+
+#ifdef WITH_CUDA
+
+#  include "device/cuda/kernel.h"
+#  include "device/cuda/device_impl.h"
+
+CCL_NAMESPACE_BEGIN
+
+void CUDADeviceKernels::load(CUDADevice *device)
+{
+  CUmodule cuModule = device->cuModule;
+
+  for (int i = 0; i < (int)DEVICE_KERNEL_NUM; i++) {
+    CUDADeviceKernel &kernel = kernels_[i];
+
+    /* No mega-kernel used for GPU. */
+    if (i == DEVICE_KERNEL_INTEGRATOR_MEGAKERNEL) {
+      continue;
+    }
+
+    const std::string function_name = std::string("kernel_gpu_") +
+                                      device_kernel_as_string((DeviceKernel)i);
+    cuda_device_assert(device,
+                       cuModuleGetFunction(&kernel.function, cuModule, function_name.c_str()));
+
+    if (kernel.function) {
+      cuda_device_assert(device, cuFuncSetCacheConfig(kernel.function, CU_FUNC_CACHE_PREFER_L1));
+
+      cuda_device_assert(
+          device,
+          cuOccupancyMaxPotentialBlockSize(
+              &kernel.min_blocks, &kernel.num_threads_per_block, kernel.function, NULL, 0, 0));
+    }
+    else {
+      LOG(ERROR) << "Unable to load kernel " << function_name;
+    }
+  }
+
+  loaded = true;
+}
+
+const CUDADeviceKernel &CUDADeviceKernels::get(DeviceKernel kernel) const
+{
+  return kernels_[(int)kernel];
+}
+
+bool CUDADeviceKernels::available(DeviceKernel kernel) const
+{
+  return kernels_[(int)kernel].function != nullptr;
+}
+
+CCL_NAMESPACE_END
+
+#endif /* WITH_CUDA*/

intern/cycles/device/cuda/kernel.h

@@ -0,0 +1,56 @@
+/*
+ * Copyright 2011-2021 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_CUDA
+
+#  include "device/device_kernel.h"
+
+#  ifdef WITH_CUDA_DYNLOAD
+#    include "cuew.h"
+#  else
+#    include <cuda.h>
+#  endif
+
+CCL_NAMESPACE_BEGIN
+
+class CUDADevice;
+
+/* CUDA kernel and associate occupancy information. */
+class CUDADeviceKernel {
+ public:
+  CUfunction function = nullptr;
+
+  int num_threads_per_block = 0;
+  int min_blocks = 0;
+};
+
+/* Cache of CUDA kernels for each DeviceKernel. */
+class CUDADeviceKernels {
+ public:
+  void load(CUDADevice *device);
+  const CUDADeviceKernel &get(DeviceKernel kernel) const;
+  bool available(DeviceKernel kernel) const;
+
+ protected:
+  CUDADeviceKernel kernels_[DEVICE_KERNEL_NUM];
+  bool loaded = false;
+};
+
+CCL_NAMESPACE_END
+
+#endif /* WITH_CUDA */

intern/cycles/device/cuda/queue.cpp

@@ -0,0 +1,230 @@
+/*
+ * Copyright 2011-2013 Blender Foundation
+ *
+ * Licensed under the Apache License, Version 2.0 (the "License");
+ * you may not use this file except in compliance with the License.
+ * You may obtain a copy of the License at
+ *
+ * http://www.apache.org/licenses/LICENSE-2.0
+ *
+ * Unless required by applicable law or agreed to in writing, software
+ * distributed under the License is distributed on an "AS IS" BASIS,
+ * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+ * See the License for the specific language governing permissions and
+ * limitations under the License.
+ */
+
+#ifdef WITH_CUDA
+
+#  include "device/cuda/queue.h"
+
+#  include "device/cuda/device_impl.h"
+#  include "device/cuda/graphics_interop.h"
+#  include "device/cuda/kernel.h"
+
+CCL_NAMESPACE_BEGIN
+
+/* CUDADeviceQueue */
+
+CUDADeviceQueue::CUDADeviceQueue(CUDADevice *device)
+    : DeviceQueue(device), cuda_device_(device), cuda_stream_(nullptr)
+{
+  const CUDAContextScope scope(cuda_device_);
+  cuda_device_assert(cuda_device_, cuStreamCreate(&cuda_stream_, CU_STREAM_NON_BLOCKING));
+}
+
+CUDADeviceQueue::~CUDADeviceQueue()
+{
+  const CUDAContextScope scope(cuda_device_);
+  cuStreamDestroy(cuda_stream_);
+}
+
+int CUDADeviceQueue::num_concurrent_states(const size_t state_size) const
+{
+  int num_states = max(cuda_device_->get_num_multiprocessors() *
+                           cuda_device_->get_max_num_threads_per_multiprocessor() * 16,
+                       1048576);
+
+  const char *factor_str = getenv("CYCLES_CONCURRENT_STATES_FACTOR");
+  if (factor_str) {
+    num_states = max((int)(num_states * atof(factor_str)), 1024);
+  }
+
+  VLOG(3) << "GPU queue concurrent states: " << num_states << ", using up to "
+          << string_human_readable_size(num_states * state_size);
+
+  return num_states;
+}
+
+int CUDADeviceQueue::num_concurrent_busy_states() const
+{
+  const int max_num_threads = cuda_device_->get_num_multiprocessors() *
+                              cuda_device_->get_max_num_threads_per_multiprocessor();
+
+  if (max_num_threads == 0) {
+    return 65536;
+  }
+
+  return 4 * max_num_threads;
+}
+
+void CUDADeviceQueue::init_execution()
+{
+  /* Synchronize all textures and memory copies before executing task. */
+  CUDAContextScope scope(cuda_device_);
+  cuda_device_->load_texture_info();
+  cuda_device_assert(cuda_device_, cuCtxSynchronize());
+
+  debug_init_execution();
+}
+
+bool CUDADeviceQueue::kernel_available(DeviceKernel kernel) const
+{
+  return cuda_device_->kernels.available(kernel);
+}
+
+bool CUDADeviceQueue::enqueue(DeviceKernel kernel, const int work_size, void *args[])
+{
+  if (cuda_device_->have_error()) {
+    return false;
+  }
+
+  debug_enqueue(kernel, work_size);
+
+  const CUDAContextScope scope(cuda_device_);
+  const CUDADeviceKernel &cuda_kernel = cuda_device_->kernels.get(kernel);
+
+  /* Compute kernel launch parameters. */
+  const int num_threads_per_block = cuda_kernel.num_threads_per_block;
+  const int num_blocks = divide_up(work_size, num_threads_per_block);
+
+  int shared_mem_bytes = 0;
+
+  switch (kernel) {
+    case DEVICE_KERNEL_INTEGRATOR_QUEUED_PATHS_ARRAY:
+    case DEVICE_KERNEL_INTEGRATOR_QUEUED_SHADOW_PATHS_ARRAY:
+    case DEVICE_KERNEL_INTEGRATOR_ACTIVE_PATHS_ARRAY:
+    case DEVICE_KERNEL_INTEGRATOR_TERMINATED_PATHS_ARRAY:
+    case DEVICE_KERNEL_INTEGRATOR_SORTED_PATHS_ARRAY:
+    case DEVICE_KERNEL_INTEGRATOR_COMPACT_PATHS_ARRAY:
+      /* See parall_active_index.h for why this amount of shared memory is needed. */
+      shared_mem_bytes = (num_threads_per_block + 1) * sizeof(int);
+      break;
+
+    default:
+      break;
+  }
+
+  /* Launch kernel. */
+  assert_success(cuLaunchKernel(cuda_kernel.function,
+                                num_blocks,
+                                1,
+                                1,
+                                num_threads_per_block,
+                                1,
+                                1,
+                                shared_mem_bytes,
+                                cuda_stream_,
+                                args,
+                                0),
+                 "enqueue");
+
+  return !(cuda_device_->have_error());
+}
+
+bool CUDADeviceQueue::synchronize()
+{
+  if (cuda_device_->have_error()) {
+    return false;
+  }
+
+  const CUDAContextScope scope(cuda_device_);
+  assert_success(cuStreamSynchronize(cuda_stream_), "synchronize");
+
+  debug_synchronize();
+
+  return !(cuda_device_->have_error());
+}
+
+void CUDADeviceQueue::zero_to_device(device_memory &mem)
+{
+  assert(mem.type != MEM_GLOBAL && mem.type != MEM_TEXTURE);
+
+  if (mem.memory_size() == 0) {
+    return;
+  }
+
+  /* Allocate on demand. */
+  if (mem.device_pointer == 0) {
+    cuda_device_->mem_alloc(mem);
+  }
+
+  /* Zero memory on device. */
+  assert(mem.device_pointer != 0);
+
+  const CUDAContextScope scope(cuda_device_);
+  assert_success(
+      cuMemsetD8Async((CUdeviceptr)mem.device_pointer, 0, mem.memory_size(), cuda_stream_),
+      "zero_to_device");
+}
+
+void CUDADeviceQueue::copy_to_device(device_memory &mem)
+{
+  assert(mem.type != MEM_GLOBAL && mem.type != MEM_TEXTURE);
+
+  if (mem.memory_size() == 0) {
+    return;
+  }
+
+  /* Allocate on demand. */
+  if (mem.device_pointer == 0) {
+    cuda_device_->mem_alloc(mem);
+  }
+
+  assert(mem.device_pointer != 0);
+  assert(mem.host_pointer != nullptr);
+
+  /* Copy memory to device. */
+  const CUDAContextScope scope(cuda_device_);
+  assert_success(
+      cuMemcpyHtoDAsync(
+          (CUdeviceptr)mem.device_pointer, mem.host_pointer, mem.memory_size(), cuda_stream_),
+      "copy_to_device");
+}
+
+void CUDADeviceQueue::copy_from_device(device_memory &mem)
+{
+  assert(mem.type != MEM_GLOBAL && mem.type != MEM_TEXTURE);
+
+  if (mem.memory_size() == 0) {
+    return;
+  }
+
+  assert(mem.device_pointer != 0);
+  assert(mem.host_pointer != nullptr);
+
+  /* Copy memory from device. */
+  const CUDAContextScope scope(cuda_device_);
+  assert_success(
+      cuMemcpyDtoHAsync(
+          mem.host_pointer, (CUdeviceptr)mem.device_pointer, mem.memory_size(), cuda_stream_),
+      "copy_from_device");
+}
+
+void CUDADeviceQueue::assert_success(CUresult result, const char *operation)
+{
+  if (result != CUDA_SUCCESS) {
+    const char *name = cuewErrorString(result);
+    cuda_device_->set_error(string_printf(
+        "%s in CUDA queue %s (%s)", name, operation, debug_active_kernels().c_str()));
+  }
+}
+
+unique_ptr<DeviceGraphicsInterop> CUDADeviceQueue::graphics_interop_create()
+{
+  return make_unique<CUDADeviceGraphicsInterop>(this);
+}
+
+CCL_NAMESPACE_END
+
+#endif /* WITH_CUDA */