Cycles: combined CPU + GPU rendering support.

CPU rendering will be restricted to a BVH2, which is not ideal for raytracing
performance but can be shared with the GPU. Decoupled volume shading will be
disabled to match GPU volume sampling.

The number of CPU rendering threads is reduced to leave one core dedicated to
each GPU. Viewport rendering will also only use GPU rendering still. So along
with the BVH2 usage, perfect scaling should not be expected.

Go to User Preferences > System to enable the CPU to render alongside the GPU.

Differential Revision: https://developer.blender.org/D2873

2 files changed, 45 insertions, 15 deletions

diff --git a/intern/cycles/blender/addon/properties.py b/intern/cycles/blender/addon/properties.py
index 7b16ef1d543..2e149527066 100644
--- a/intern/cycles/blender/addon/properties.py
+++ b/intern/cycles/blender/addon/properties.py
@@ -1351,8 +1351,9 @@ class CyclesPreferences(bpy.types.AddonPreferences):
 
         cuda_devices = []
         opencl_devices = []
+        cpu_devices = []
         for device in device_list:
-            if not device[1] in {'CUDA', 'OPENCL'}:
+            if not device[1] in {'CUDA', 'OPENCL', 'CPU'}:
                 continue
 
             entry = None
@@ -1361,18 +1362,28 @@ class CyclesPreferences(bpy.types.AddonPreferences):
                 if dev.id == device[2] and dev.type == device[1]:
                     entry = dev
                     break
-            # Create new entry if no existing one was found
             if not entry:
+                # Create new entry if no existing one was found
                 entry = self.devices.add()
                 entry.id   = device[2]
                 entry.name = device[0]
                 entry.type = device[1]
+                entry.use  = entry.type != 'CPU'
+            elif entry.name != device[0]:
+                # Update name in case it changed
+                entry.name = device[0]
 
             # Sort entries into lists
             if entry.type == 'CUDA':
                 cuda_devices.append(entry)
             elif entry.type == 'OPENCL':
                 opencl_devices.append(entry)
+            else:
+                cpu_devices.append(entry)
+
+        cuda_devices.extend(cpu_devices)
+        opencl_devices.extend(cpu_devices)
+
         return cuda_devices, opencl_devices
 
 
diff --git a/intern/cycles/blender/blender_sync.cpp b/intern/cycles/blender/blender_sync.cpp
index 2e3301c4209..5eddf189468 100644
--- a/intern/cycles/blender/blender_sync.cpp
+++ b/intern/cycles/blender/blender_sync.cpp
@@ -675,6 +675,15 @@ SessionParams BlenderSync::get_session_params(BL::RenderEngine& b_engine,
 	/* feature set */
 	params.experimental = (get_enum(cscene, "feature_set") != 0);
 
+	/* threads */
+	if(b_scene.render().threads_mode() == BL::RenderSettings::threads_mode_FIXED)
+		params.threads = b_scene.render().threads();
+	else
+		params.threads = 0;
+
+	/* Background */
+	params.background = background;
+
 	/* device type */
 	vector<DeviceInfo>& devices = Device::available_devices();
 	
@@ -703,12 +712,28 @@ SessionParams BlenderSync::get_session_params(BL::RenderEngine& b_engine,
 			}
 		}
 
-		int compute_device = get_enum(b_preferences, "compute_device_type");
+		enum ComputeDevice {
+			COMPUTE_DEVICE_CPU = 0,
+			COMPUTE_DEVICE_CUDA = 1,
+			COMPUTE_DEVICE_OPENCL = 2,
+			COMPUTE_DEVICE_NUM = 3,
+		};
 
-		if(compute_device != 0) {
+		ComputeDevice compute_device = (ComputeDevice)get_enum(b_preferences,
+		                                                       "compute_device_type",
+		                                                       COMPUTE_DEVICE_NUM,
+		                                                       COMPUTE_DEVICE_CPU);
+
+		if(compute_device != COMPUTE_DEVICE_CPU) {
 			vector<DeviceInfo> used_devices;
 			RNA_BEGIN(&b_preferences, device, "devices") {
-				if(get_enum(device, "type") == compute_device && get_boolean(device, "use")) {
+				ComputeDevice device_type = (ComputeDevice)get_enum(device,
+				                                                    "type",
+				                                                    COMPUTE_DEVICE_NUM,
+				                                                    COMPUTE_DEVICE_CPU);
+
+				if(get_boolean(device, "use") &&
+				   (device_type == compute_device || device_type == COMPUTE_DEVICE_CPU)) {
 					string id = get_string(device, "id");
 					foreach(DeviceInfo& info, devices) {
 						if(info.id == id) {
@@ -723,15 +748,14 @@ SessionParams BlenderSync::get_session_params(BL::RenderEngine& b_engine,
 				params.device = used_devices[0];
 			}
 			else if(used_devices.size() > 1) {
-				params.device = Device::get_multi_device(used_devices);
+				params.device = Device::get_multi_device(used_devices,
+				                                         params.threads,
+				                                         params.background);
 			}
 			/* Else keep using the CPU device that was set before. */
 		}
 	}
 
-	/* Background */
-	params.background = background;
-
 	/* samples */
 	int samples = get_int(cscene, "samples");
 	int aa_samples = get_int(cscene, "aa_samples");
@@ -791,15 +815,10 @@ SessionParams BlenderSync::get_session_params(BL::RenderEngine& b_engine,
 		params.tile_order = TILE_BOTTOM_TO_TOP;
 	}
 
+	/* other parameters */
 	params.start_resolution = get_int(cscene, "preview_start_resolution");
 	params.pixel_size = b_engine.get_preview_pixel_size(b_scene);
 
-	/* other parameters */
-	if(b_scene.render().threads_mode() == BL::RenderSettings::threads_mode_FIXED)
-		params.threads = b_scene.render().threads();
-	else
-		params.threads = 0;
-
 	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");