Merge branch 'master' into blender2.8

6 files changed, 57 insertions, 29 deletions

diff --git a/intern/cycles/device/device.cpp b/intern/cycles/device/device.cpp
index 5ae83b56fcd..19c4bec55a8 100644
--- a/intern/cycles/device/device.cpp
+++ b/intern/cycles/device/device.cpp
@@ -35,6 +35,7 @@ CCL_NAMESPACE_BEGIN
 
 bool Device::need_types_update = true;
 bool Device::need_devices_update = true;
+thread_mutex Device::device_mutex;
 vector<DeviceType> Device::types;
 vector<DeviceInfo> Device::devices;
 
@@ -443,53 +444,49 @@ string Device::string_from_type(DeviceType type)
 
 vector<DeviceType>& Device::available_types()
 {
+	thread_scoped_lock lock(device_mutex);
 	if(need_types_update) {
 		types.clear();
 		types.push_back(DEVICE_CPU);
-
 #ifdef WITH_CUDA
-		if(device_cuda_init())
+		if(device_cuda_init()) {
 			types.push_back(DEVICE_CUDA);
+		}
 #endif
-
 #ifdef WITH_OPENCL
-		if(device_opencl_init())
+		if(device_opencl_init()) {
 			types.push_back(DEVICE_OPENCL);
+		}
 #endif
-
 #ifdef WITH_NETWORK
 		types.push_back(DEVICE_NETWORK);
 #endif
-
 		need_types_update = false;
 	}
-
 	return types;
 }
 
 vector<DeviceInfo>& Device::available_devices()
 {
+	thread_scoped_lock lock(device_mutex);
 	if(need_devices_update) {
 		devices.clear();
-#ifdef WITH_CUDA
-		if(device_cuda_init())
-			device_cuda_info(devices);
-#endif
-
 #ifdef WITH_OPENCL
-		if(device_opencl_init())
+		if(device_opencl_init()) {
 			device_opencl_info(devices);
+		}
+#endif
+#ifdef WITH_CUDA
+		if(device_cuda_init()) {
+			device_cuda_info(devices);
+		}
 #endif
-
 		device_cpu_info(devices);
-
 #ifdef WITH_NETWORK
 		device_network_info(devices);
 #endif
-
 		need_devices_update = false;
 	}
-
 	return devices;
 }
 
@@ -497,12 +494,6 @@ string Device::device_capabilities()
 {
 	string capabilities = "CPU device capabilities: ";
 	capabilities += device_cpu_capabilities() + "\n";
-#ifdef WITH_CUDA
-	if(device_cuda_init()) {
-		capabilities += "\nCUDA device capabilities:\n";
-		capabilities += device_cuda_capabilities();
-	}
-#endif
 
 #ifdef WITH_OPENCL
 	if(device_opencl_init()) {
@@ -511,6 +502,13 @@ string Device::device_capabilities()
 	}
 #endif
 
+#ifdef WITH_CUDA
+	if(device_cuda_init()) {
+		capabilities += "\nCUDA device capabilities:\n";
+		capabilities += device_cuda_capabilities();
+	}
+#endif
+
 	return capabilities;
 }
 
@@ -526,10 +524,14 @@ DeviceInfo Device::get_multi_device(vector<DeviceInfo> subdevices)
 	info.num = 0;
 
 	info.has_bindless_textures = true;
+	info.has_volume_decoupled = true;
+	info.has_qbvh = true;
 	foreach(DeviceInfo &device, subdevices) {
 		assert(device.type == info.multi_devices[0].type);
 
 		info.has_bindless_textures &= device.has_bindless_textures;
+		info.has_volume_decoupled &= device.has_volume_decoupled;
+		info.has_qbvh &= device.has_qbvh;
 	}
 
 	return info;
diff --git a/intern/cycles/device/device.h b/intern/cycles/device/device.h
index fe0bcc5b91f..29803abd153 100644
--- a/intern/cycles/device/device.h
+++ b/intern/cycles/device/device.h
@@ -55,6 +55,8 @@ public:
 	bool display_device;
 	bool advanced_shading;
 	bool has_bindless_textures; /* flag for GPU and Multi device */
+	bool has_volume_decoupled;
+	bool has_qbvh;
 	bool use_split_kernel; /* Denotes if the device is going to run cycles using split-kernel */
 	vector<DeviceInfo> multi_devices;
 
@@ -66,6 +68,8 @@ public:
 		display_device = false;
 		advanced_shading = true;
 		has_bindless_textures = false;
+		has_volume_decoupled = false;
+		has_qbvh = false;
 		use_split_kernel = false;
 	}
 
@@ -364,6 +368,7 @@ public:
 private:
 	/* Indicted whether device types and devices lists were initialized. */
 	static bool need_types_update, need_devices_update;
+	static thread_mutex device_mutex;
 	static vector<DeviceType> types;
 	static vector<DeviceInfo> devices;
 };
diff --git a/intern/cycles/device/device_cpu.cpp b/intern/cycles/device/device_cpu.cpp
index ac6d3246d38..a17caabc850 100644
--- a/intern/cycles/device/device_cpu.cpp
+++ b/intern/cycles/device/device_cpu.cpp
@@ -1024,6 +1024,8 @@ void device_cpu_info(vector<DeviceInfo>& devices)
 	info.id = "CPU";
 	info.num = 0;
 	info.advanced_shading = true;
+	info.has_qbvh = system_cpu_support_sse2();
+	info.has_volume_decoupled = true;
 
 	devices.insert(devices.begin(), info);
 }
diff --git a/intern/cycles/device/device_cuda.cpp b/intern/cycles/device/device_cuda.cpp
index 3d209e5560c..066be82d55b 100644
--- a/intern/cycles/device/device_cuda.cpp
+++ b/intern/cycles/device/device_cuda.cpp
@@ -1313,9 +1313,14 @@ public:
 		CUdeviceptr d_work_tiles = cuda_device_ptr(work_tiles.device_pointer);
 
 		/* Prepare work size. More step samples render faster, but for now we
-		 * remain conservative to avoid driver timeouts. */
+		 * remain conservative for GPUs connected to a display to avoid driver
+		 * timeouts and display freezing. */
 		int min_blocks, num_threads_per_block;
 		cuda_assert(cuOccupancyMaxPotentialBlockSize(&min_blocks, &num_threads_per_block, cuPathTrace, NULL, 0, 0));
+		if(!info.display_device) {
+			min_blocks *= 8;
+		}
+
 		uint step_samples = divide_up(min_blocks * num_threads_per_block, wtile->w * wtile->h);;
 
 		/* Render all samples. */
@@ -2130,7 +2135,6 @@ void device_cuda_info(vector<DeviceInfo>& devices)
 
 	for(int num = 0; num < count; num++) {
 		char name[256];
-		int attr;
 
 		if(cuDeviceGetName(name, 256, num) != CUDA_SUCCESS)
 			continue;
@@ -2149,6 +2153,8 @@ void device_cuda_info(vector<DeviceInfo>& devices)
 
 		info.advanced_shading = (major >= 2);
 		info.has_bindless_textures = (major >= 3);
+		info.has_volume_decoupled = false;
+		info.has_qbvh = false;
 
 		int pci_location[3] = {0, 0, 0};
 		cuDeviceGetAttribute(&pci_location[0], CU_DEVICE_ATTRIBUTE_PCI_DOMAIN_ID, num);
@@ -2160,14 +2166,21 @@ void device_cuda_info(vector<DeviceInfo>& devices)
 		                        (unsigned int)pci_location[1],
 		                        (unsigned int)pci_location[2]);
 
-		/* if device has a kernel timeout, assume it is used for display */
-		if(cuDeviceGetAttribute(&attr, CU_DEVICE_ATTRIBUTE_KERNEL_EXEC_TIMEOUT, num) == CUDA_SUCCESS && attr == 1) {
+		/* If device has a kernel timeout and no compute preemption, we assume
+		 * it is connected to a display and will freeze the display while doing
+		 * computations. */
+		int timeout_attr = 0, preempt_attr = 0;
+		cuDeviceGetAttribute(&timeout_attr, CU_DEVICE_ATTRIBUTE_KERNEL_EXEC_TIMEOUT, num);
+		cuDeviceGetAttribute(&preempt_attr, CU_DEVICE_ATTRIBUTE_COMPUTE_PREEMPTION_SUPPORTED, num);
+
+		if(timeout_attr && !preempt_attr) {
 			info.description += " (Display)";
 			info.display_device = true;
 			display_devices.push_back(info);
 		}
-		else
+		else {
 			devices.push_back(info);
+		}
 	}
 
 	if(!display_devices.empty())
diff --git a/intern/cycles/device/device_network.cpp b/intern/cycles/device/device_network.cpp
index deea59f1d23..ced10c98dc9 100644
--- a/intern/cycles/device/device_network.cpp
+++ b/intern/cycles/device/device_network.cpp
@@ -343,7 +343,11 @@ void device_network_info(vector<DeviceInfo>& devices)
 	info.description = "Network Device";
 	info.id = "NETWORK";
 	info.num = 0;
-	info.advanced_shading = true; /* todo: get this info from device */
+
+	/* todo: get this info from device */
+	info.advanced_shading = true;
+	info.has_volume_decoupled = false;
+	info.has_qbvh = false;
 
 	devices.push_back(info);
 }
diff --git a/intern/cycles/device/device_opencl.cpp b/intern/cycles/device/device_opencl.cpp
index 9d89decaaaf..5808a31e605 100644
--- a/intern/cycles/device/device_opencl.cpp
+++ b/intern/cycles/device/device_opencl.cpp
@@ -123,6 +123,8 @@ void device_opencl_info(vector<DeviceInfo>& devices)
 		info.advanced_shading = OpenCLInfo::kernel_use_advanced_shading(platform_name);
 		info.use_split_kernel = OpenCLInfo::kernel_use_split(platform_name,
 		                                                     device_type);
+		info.has_volume_decoupled = false;
+		info.has_qbvh = false;
 		info.id = string("OPENCL_") + platform_name + "_" + device_name + "_" + hardware_id;
 		devices.push_back(info);
 		num_devices++;