Merge pull request #100062 from thimenesup/rd_buffer_address

Implement Buffer Device Address for Rendering Device Vulkan and DirectX12

Author: akien-mga

doc/classes/RenderingDevice.xml

@@ -81,6 +81,14 @@
 				[/codeblock]
 			</description>
 		</method>
+		<method name="buffer_get_device_address">
+			<return type="int" />
+			<param index="0" name="buffer" type="RID" />
+			<description>
+				Returns the address of the given [param buffer] which can be passed to shaders in any way to access underlying data. Buffer must have been created with this feature enabled.
+				[b]Note:[/b] You must check that the GPU supports this functionality by calling [method has_feature] with [constant SUPPORTS_BUFFER_DEVICE_ADDRESS] as a parameter.
+			</description>
+		</method>
 		<method name="buffer_update">
 			<return type="int" enum="Error" />
 			<param index="0" name="buffer" type="RID" />
@@ -672,6 +680,13 @@
 				This is only used by Vulkan in debug builds. Godot must also be started with the [code]--extra-gpu-memory-tracking[/code] [url=$DOCS_URL/tutorials/editor/command_line_tutorial.html]command line argument[/url].
 			</description>
 		</method>
+		<method name="has_feature" qualifiers="const">
+			<return type="bool" />
+			<param index="0" name="feature" type="int" enum="RenderingDevice.Features" />
+			<description>
+				Returns [code]true[/code] if the [param feature] is supported by the GPU.
+			</description>
+		</method>
 		<method name="index_array_create">
 			<return type="RID" />
 			<param index="0" name="index_buffer" type="RID" />
@@ -688,9 +703,11 @@
 			<param index="1" name="format" type="int" enum="RenderingDevice.IndexBufferFormat" />
 			<param index="2" name="data" type="PackedByteArray" default="PackedByteArray()" />
 			<param index="3" name="use_restart_indices" type="bool" default="false" />
+			<param index="4" name="enable_device_address" type="bool" default="false" />
 			<description>
 				Creates a new index buffer. It can be accessed with the RID that is returned.
 				Once finished with your RID, you will want to free the RID using the RenderingDevice's [method free_rid] method.
+				Optionally, set [param enable_device_address] if you wish to use [method buffer_get_device_address] functionality and the GPU supports it.
 			</description>
 		</method>
 		<method name="limit_get" qualifiers="const">
@@ -1056,9 +1073,11 @@
 			<return type="RID" />
 			<param index="0" name="size_bytes" type="int" />
 			<param index="1" name="data" type="PackedByteArray" default="PackedByteArray()" />
+			<param index="2" name="enable_device_address" type="bool" default="false" />
 			<description>
 				Creates a new uniform buffer. It can be accessed with the RID that is returned.
 				Once finished with your RID, you will want to free the RID using the RenderingDevice's [method free_rid] method.
+				Optionally, set [param enable_device_address] if you wish to use [method buffer_get_device_address] functionality and the GPU supports it.
 			</description>
 		</method>
 		<method name="uniform_set_create">
@@ -1093,9 +1112,11 @@
 			<param index="0" name="size_bytes" type="int" />
 			<param index="1" name="data" type="PackedByteArray" default="PackedByteArray()" />
 			<param index="2" name="use_as_storage" type="bool" default="false" />
+			<param index="3" name="enable_device_address" type="bool" default="false" />
 			<description>
 				It can be accessed with the RID that is returned.
 				Once finished with your RID, you will want to free the RID using the RenderingDevice's [method free_rid] method.
+				Optionally, set [param enable_device_address] if you wish to use [method buffer_get_device_address] functionality and the GPU supports it.
 			</description>
 		</method>
 		<method name="vertex_format_create">
@@ -2047,6 +2068,9 @@
 		</constant>
 		<constant name="STORAGE_BUFFER_USAGE_DISPATCH_INDIRECT" value="1" enum="StorageBufferUsage" is_bitfield="true">
 		</constant>
+		<constant name="STORAGE_BUFFER_USAGE_DEVICE_ADDRESS" value="2" enum="StorageBufferUsage" is_bitfield="true">
+			Allows usage of [method buffer_get_device_address] on supported GPUs.
+		</constant>
 		<constant name="UNIFORM_TYPE_SAMPLER" value="0" enum="UniformType">
 			Sampler uniform.
 		</constant>
@@ -2418,6 +2442,9 @@
 		<constant name="PIPELINE_SPECIALIZATION_CONSTANT_TYPE_FLOAT" value="2" enum="PipelineSpecializationConstantType">
 			Floating-point specialization constant.
 		</constant>
+		<constant name="SUPPORTS_BUFFER_DEVICE_ADDRESS" value="6" enum="Features">
+			Features support for buffer device address extension.
+		</constant>
 		<constant name="LIMIT_MAX_BOUND_UNIFORM_SETS" value="0" enum="Limit">
 			Maximum number of uniform sets that can be bound at a given time.
 		</constant>

drivers/d3d12/rendering_device_driver_d3d12.cpp

@@ -905,6 +905,11 @@ void RenderingDeviceDriverD3D12::buffer_unmap(BufferID p_buffer) {
 	buf_info->resource->Unmap(0, &VOID_RANGE);
 }
 
+uint64_t RenderingDeviceDriverD3D12::buffer_get_device_address(BufferID p_buffer) {
+	const BufferInfo *buf_info = (const BufferInfo *)p_buffer.id;
+	return buf_info->resource->GetGPUVirtualAddress();
+}
+
 /*****************/
 /**** TEXTURE ****/
 /*****************/
@@ -6271,6 +6276,8 @@ bool RenderingDeviceDriverD3D12::has_feature(Features p_feature) {
 			return vrs_capabilities.ss_image_supported;
 		case SUPPORTS_FRAGMENT_SHADER_WITH_ONLY_SIDE_EFFECTS:
 			return true;
+		case SUPPORTS_BUFFER_DEVICE_ADDRESS:
+			return true;
 		default:
 			return false;
 	}

drivers/d3d12/rendering_device_driver_d3d12.h

@@ -284,6 +284,7 @@ class RenderingDeviceDriverD3D12 : public RenderingDeviceDriver {
 	virtual uint64_t buffer_get_allocation_size(BufferID p_buffer) override final;
 	virtual uint8_t *buffer_map(BufferID p_buffer) override final;
 	virtual void buffer_unmap(BufferID p_buffer) override final;
+	virtual uint64_t buffer_get_device_address(BufferID p_buffer) override final;
 
 	/*****************/
 	/**** TEXTURE ****/

drivers/metal/rendering_device_driver_metal.h

@@ -105,6 +105,7 @@ class API_AVAILABLE(macos(11.0), ios(14.0), tvos(14.0)) RenderingDeviceDriverMet
 	virtual uint64_t buffer_get_allocation_size(BufferID p_buffer) override final;
 	virtual uint8_t *buffer_map(BufferID p_buffer) override final;
 	virtual void buffer_unmap(BufferID p_buffer) override final;
+	virtual uint64_t buffer_get_device_address(BufferID p_buffer) override final;
 
 #pragma mark - Texture
 

drivers/metal/rendering_device_driver_metal.mm

@@ -159,6 +159,15 @@ _FORCE_INLINE_ MTLSize mipmapLevelSizeFromSize(MTLSize p_size, NSUInteger p_leve
 	// Nothing to do.
 }
 
+uint64_t RenderingDeviceDriverMetal::buffer_get_device_address(BufferID p_buffer) {
+	if (@available(iOS 16.0, macOS 13.0, *)) {
+		id<MTLBuffer> obj = rid::get(p_buffer);
+		return obj.gpuAddress;
+	} else {
+		return 0;
+	}
+}
+
 #pragma mark - Texture
 
 #pragma mark - Format Conversions
@@ -4026,6 +4035,8 @@ bool isArrayTexture(MTLTextureType p_type) {
 			return false;
 		case SUPPORTS_FRAGMENT_SHADER_WITH_ONLY_SIDE_EFFECTS:
 			return true;
+		case SUPPORTS_BUFFER_DEVICE_ADDRESS:
+			return false;
 		case SUPPORTS_METALFX_SPATIAL:
 			return device_properties->features.metal_fx_spatial;
 		case SUPPORTS_METALFX_TEMPORAL:

drivers/vulkan/rendering_device_driver_vulkan.cpp

@@ -513,6 +513,7 @@ Error RenderingDeviceDriverVulkan::_initialize_device_extensions() {
 	_register_requested_device_extension(VK_EXT_PIPELINE_CREATION_CACHE_CONTROL_EXTENSION_NAME, false);
 	_register_requested_device_extension(VK_EXT_SUBGROUP_SIZE_CONTROL_EXTENSION_NAME, false);
 	_register_requested_device_extension(VK_EXT_ASTC_DECODE_MODE_EXTENSION_NAME, false);
+	_register_requested_device_extension(VK_KHR_BUFFER_DEVICE_ADDRESS_EXTENSION_NAME, false);
 
 	if (Engine::get_singleton()->is_generate_spirv_debug_info_enabled()) {
 		_register_requested_device_extension(VK_KHR_SHADER_NON_SEMANTIC_INFO_EXTENSION_NAME, true);
@@ -730,6 +731,7 @@ Error RenderingDeviceDriverVulkan::_check_device_capabilities() {
 		void *next_features = nullptr;
 		VkPhysicalDeviceVulkan12Features device_features_vk_1_2 = {};
 		VkPhysicalDeviceShaderFloat16Int8FeaturesKHR shader_features = {};
+		VkPhysicalDeviceBufferDeviceAddressFeaturesKHR buffer_device_address_features = {};
 		VkPhysicalDeviceFragmentShadingRateFeaturesKHR vrs_features = {};
 		VkPhysicalDevice16BitStorageFeaturesKHR storage_feature = {};
 		VkPhysicalDeviceMultiviewFeatures multiview_features = {};
@@ -740,10 +742,17 @@ Error RenderingDeviceDriverVulkan::_check_device_capabilities() {
 			device_features_vk_1_2.sType = VK_STRUCTURE_TYPE_PHYSICAL_DEVICE_VULKAN_1_2_FEATURES;
 			device_features_vk_1_2.pNext = next_features;
 			next_features = &device_features_vk_1_2;
-		} else if (enabled_device_extension_names.has(VK_KHR_SHADER_FLOAT16_INT8_EXTENSION_NAME)) {
-			shader_features.sType = VK_STRUCTURE_TYPE_PHYSICAL_DEVICE_SHADER_FLOAT16_INT8_FEATURES_KHR;
-			shader_features.pNext = next_features;
-			next_features = &shader_features;
+		} else {
+			if (enabled_device_extension_names.has(VK_KHR_SHADER_FLOAT16_INT8_EXTENSION_NAME)) {
+				shader_features.sType = VK_STRUCTURE_TYPE_PHYSICAL_DEVICE_SHADER_FLOAT16_INT8_FEATURES_KHR;
+				shader_features.pNext = next_features;
+				next_features = &shader_features;
+			}
+			if (enabled_device_extension_names.has(VK_KHR_BUFFER_DEVICE_ADDRESS_EXTENSION_NAME)) {
+				buffer_device_address_features.sType = VK_STRUCTURE_TYPE_PHYSICAL_DEVICE_BUFFER_DEVICE_ADDRESS_FEATURES_KHR;
+				buffer_device_address_features.pNext = next_features;
+				next_features = &buffer_device_address_features;
+			}
 		}
 
 		if (enabled_device_extension_names.has(VK_KHR_FRAGMENT_SHADING_RATE_EXTENSION_NAME)) {
@@ -783,11 +792,17 @@ Error RenderingDeviceDriverVulkan::_check_device_capabilities() {
 				shader_capabilities.shader_float16_is_supported = device_features_vk_1_2.shaderFloat16;
 				shader_capabilities.shader_int8_is_supported = device_features_vk_1_2.shaderInt8;
 			}
+			if (enabled_device_extension_names.has(VK_KHR_BUFFER_DEVICE_ADDRESS_EXTENSION_NAME)) {
+				buffer_device_address_support = device_features_vk_1_2.bufferDeviceAddress;
+			}
 		} else {
 			if (enabled_device_extension_names.has(VK_KHR_SHADER_FLOAT16_INT8_EXTENSION_NAME)) {
 				shader_capabilities.shader_float16_is_supported = shader_features.shaderFloat16;
 				shader_capabilities.shader_int8_is_supported = shader_features.shaderInt8;
 			}
+			if (enabled_device_extension_names.has(VK_KHR_BUFFER_DEVICE_ADDRESS_EXTENSION_NAME)) {
+				buffer_device_address_support = buffer_device_address_features.bufferDeviceAddress;
+			}
 		}
 
 		if (enabled_device_extension_names.has(VK_KHR_FRAGMENT_SHADING_RATE_EXTENSION_NAME)) {
@@ -971,6 +986,14 @@ Error RenderingDeviceDriverVulkan::_initialize_device(const LocalVector<VkDevice
 	shader_features.shaderInt8 = shader_capabilities.shader_int8_is_supported;
 	create_info_next = &shader_features;
 
+	VkPhysicalDeviceBufferDeviceAddressFeaturesKHR buffer_device_address_features = {};
+	if (buffer_device_address_support) {
+		buffer_device_address_features.sType = VK_STRUCTURE_TYPE_PHYSICAL_DEVICE_BUFFER_DEVICE_ADDRESS_FEATURES_KHR;
+		buffer_device_address_features.pNext = create_info_next;
+		buffer_device_address_features.bufferDeviceAddress = buffer_device_address_support;
+		create_info_next = &buffer_device_address_features;
+	}
+
 	VkPhysicalDeviceFragmentShadingRateFeaturesKHR vrs_features = {};
 	if (vrs_capabilities.pipeline_vrs_supported || vrs_capabilities.primitive_vrs_supported || vrs_capabilities.attachment_vrs_supported) {
 		vrs_features.sType = VK_STRUCTURE_TYPE_PHYSICAL_DEVICE_FRAGMENT_SHADING_RATE_FEATURES_KHR;
@@ -1112,6 +1135,9 @@ Error RenderingDeviceDriverVulkan::_initialize_allocator() {
 	if (use_1_3_features) {
 		allocator_info.flags |= VMA_ALLOCATOR_CREATE_KHR_MAINTENANCE5_BIT;
 	}
+	if (buffer_device_address_support) {
+		allocator_info.flags |= VMA_ALLOCATOR_CREATE_BUFFER_DEVICE_ADDRESS_BIT;
+	}
 	VkResult err = vmaCreateAllocator(&allocator_info, &allocator);
 	ERR_FAIL_COND_V_MSG(err, ERR_CANT_CREATE, "vmaCreateAllocator failed with error " + itos(err) + ".");
 
@@ -1487,6 +1513,7 @@ static_assert(ENUM_MEMBERS_EQUAL(RDD::BUFFER_USAGE_STORAGE_BIT, VK_BUFFER_USAGE_
 static_assert(ENUM_MEMBERS_EQUAL(RDD::BUFFER_USAGE_INDEX_BIT, VK_BUFFER_USAGE_INDEX_BUFFER_BIT));
 static_assert(ENUM_MEMBERS_EQUAL(RDD::BUFFER_USAGE_VERTEX_BIT, VK_BUFFER_USAGE_VERTEX_BUFFER_BIT));
 static_assert(ENUM_MEMBERS_EQUAL(RDD::BUFFER_USAGE_INDIRECT_BIT, VK_BUFFER_USAGE_INDIRECT_BUFFER_BIT));
+static_assert(ENUM_MEMBERS_EQUAL(RDD::BUFFER_USAGE_DEVICE_ADDRESS_BIT, VK_BUFFER_USAGE_SHADER_DEVICE_ADDRESS_BIT));
 
 RDD::BufferID RenderingDeviceDriverVulkan::buffer_create(uint64_t p_size, BitField<BufferUsageBits> p_usage, MemoryAllocationType p_allocation_type) {
 	VkBufferCreateInfo create_info = {};
@@ -1588,6 +1615,15 @@ void RenderingDeviceDriverVulkan::buffer_unmap(BufferID p_buffer) {
 	vmaUnmapMemory(allocator, buf_info->allocation.handle);
 }
 
+uint64_t RenderingDeviceDriverVulkan::buffer_get_device_address(BufferID p_buffer) {
+	const BufferInfo *buf_info = (const BufferInfo *)p_buffer.id;
+	VkBufferDeviceAddressInfo address_info = {};
+	address_info.sType = VK_STRUCTURE_TYPE_BUFFER_DEVICE_ADDRESS_INFO;
+	address_info.pNext = nullptr;
+	address_info.buffer = buf_info->vk_buffer;
+	return vkGetBufferDeviceAddress(vk_device, &address_info);
+}
+
 /*****************/
 /**** TEXTURE ****/
 /*****************/
@@ -5874,6 +5910,8 @@ bool RenderingDeviceDriverVulkan::has_feature(Features p_feature) {
 			return vrs_capabilities.attachment_vrs_supported && physical_device_features.shaderStorageImageExtendedFormats;
 		case SUPPORTS_FRAGMENT_SHADER_WITH_ONLY_SIDE_EFFECTS:
 			return true;
+		case SUPPORTS_BUFFER_DEVICE_ADDRESS:
+			return buffer_device_address_support;
 		default:
 			return false;
 	}

drivers/vulkan/rendering_device_driver_vulkan.h

@@ -138,6 +138,7 @@ class RenderingDeviceDriverVulkan : public RenderingDeviceDriver {
 	VRSCapabilities vrs_capabilities;
 	ShaderCapabilities shader_capabilities;
 	StorageBufferCapabilities storage_buffer_capabilities;
+	bool buffer_device_address_support = false;
 	bool pipeline_cache_control_support = false;
 	bool device_fault_support = false;
 #if defined(VK_TRACK_DEVICE_MEMORY)
@@ -204,6 +205,7 @@ class RenderingDeviceDriverVulkan : public RenderingDeviceDriver {
 	virtual uint64_t buffer_get_allocation_size(BufferID p_buffer) override final;
 	virtual uint8_t *buffer_map(BufferID p_buffer) override final;
 	virtual void buffer_unmap(BufferID p_buffer) override final;
+	virtual uint64_t buffer_get_device_address(BufferID p_buffer) override final;
 
 	/*****************/
 	/**** TEXTURE ****/

misc/extension_api_validation/4.3-stable.expected

@@ -285,3 +285,12 @@ GH-101482
 Validate extension JSON: Error: Field 'classes/RichTextLabel/methods/set_table_column_expand/arguments': size changed value in new API, from 3 to 4.
 
 Added optional "shrink" argument. Compatibility method registered.
+
+
+GH-100062
+--------
+Validate extension JSON: Error: Field 'classes/RenderingDevice/methods/index_buffer_create/arguments': size changed value in new API, from 4 to 5.
+Validate extension JSON: Error: Field 'classes/RenderingDevice/methods/uniform_buffer_create/arguments': size changed value in new API, from 2 to 3.
+Validate extension JSON: Error: Field 'classes/RenderingDevice/methods/vertex_buffer_create/arguments': size changed value in new API, from 3 to 4.
+
+Optional argument added. Compatibility methods registered.

servers/rendering/rendering_device.compat.inc

@@ -143,6 +143,18 @@ RenderingDevice::FramebufferFormatID RenderingDevice::_screen_get_framebuffer_fo
 	return screen_get_framebuffer_format(DisplayServer::MAIN_WINDOW_ID);
 }
 
+RID RenderingDevice::_uniform_buffer_create_bind_compat_100062(uint32_t p_size_bytes, const Vector<uint8_t> &p_data) {
+	return uniform_buffer_create(p_size_bytes, p_data, false);
+}
+
+RID RenderingDevice::_vertex_buffer_create_bind_compat_100062(uint32_t p_size_bytes, const Vector<uint8_t> &p_data, bool p_use_as_storage) {
+	return vertex_buffer_create(p_size_bytes, p_data, p_use_as_storage, false);
+}
+
+RID RenderingDevice::_index_buffer_create_bind_compat_100062(uint32_t p_size_indices, IndexBufferFormat p_format, const Vector<uint8_t> &p_data, bool p_use_restart_indices) {
+	return index_buffer_create(p_size_indices, p_format, p_data, p_use_restart_indices, false);
+}
+
 void RenderingDevice::_bind_compatibility_methods() {
 	ClassDB::bind_compatibility_method(D_METHOD("shader_create_from_bytecode", "binary_data"), &RenderingDevice::_shader_create_from_bytecode_bind_compat_79606);
 
@@ -166,6 +178,10 @@ void RenderingDevice::_bind_compatibility_methods() {
 	ClassDB::bind_compatibility_method(D_METHOD("draw_list_begin", "framebuffer", "initial_color_action", "final_color_action", "initial_depth_action", "final_depth_action", "clear_color_values", "clear_depth", "clear_stencil", "region"), &RenderingDevice::_draw_list_begin_bind_compat_90993, DEFVAL(Vector<Color>()), DEFVAL(1.0), DEFVAL(0), DEFVAL(Rect2()));
 
 	ClassDB::bind_compatibility_method(D_METHOD("draw_list_begin", "framebuffer", "initial_color_action", "final_color_action", "initial_depth_action", "final_depth_action", "clear_color_values", "clear_depth", "clear_stencil", "region", "breadcrumb"), &RenderingDevice::_draw_list_begin_bind_compat_98670, DEFVAL(Vector<Color>()), DEFVAL(1.0), DEFVAL(0), DEFVAL(Rect2()), DEFVAL(0));
+
+	ClassDB::bind_compatibility_method(D_METHOD("uniform_buffer_create"), &RenderingDevice::_uniform_buffer_create_bind_compat_100062, DEFVAL(Vector<uint8_t>()));
+	ClassDB::bind_compatibility_method(D_METHOD("vertex_buffer_create"), &RenderingDevice::_vertex_buffer_create_bind_compat_100062, DEFVAL(Vector<uint8_t>()), DEFVAL(false));
+	ClassDB::bind_compatibility_method(D_METHOD("index_buffer_create"), &RenderingDevice::_index_buffer_create_bind_compat_100062, DEFVAL(Vector<uint8_t>()), DEFVAL(false));
 }
 
 #endif