Skip to content

Latest commit

 

History

History
2292 lines (1868 loc) · 113 KB

LoaderAndLayerInterface.md

File metadata and controls

2292 lines (1868 loc) · 113 KB

Architecture of the Vulkan Loader Interfaces

Table of Contents

Overview

Vulkan is a layered architecture, made up of the following elements:

High Level View of Loader

The general concepts in this document are applicable to the loaders available for Windows, Linux and Android based systems.

Who Should Read This Document

While this document is primarily targeted at developers of Vulkan applications, drivers and layers, the information contained in it could be useful to anyone wanting a better understanding of the Vulkan runtime.

The Loader

The application sits on one end of, and interfaces directly with, the loader. On the other end of the loader from the application are the ICDs, which control the Vulkan-capable hardware. An important point to remember is that Vulkan-capable hardware can be graphics-based, compute-based, or both. Between the application and the ICDs the loader can inject a number of optional layers that provide special functionality.

The loader is responsible for working with the various layers as well as supporting multiple GPUs and their drivers. Any Vulkan function may wind up calling into a diverse set of modules: loader, layers, and ICDs. The loader is critical to managing the proper dispatching of Vulkan functions to the appropriate set of layers and ICDs. The Vulkan object model allows the loader to insert layers into a call chain so that the layers can process Vulkan functions prior to the ICD being called.

This document is intended to provide an overview of the necessary interfaces between each of these.

Goals of the Loader

The loader was designed with the following goals in mind.

  1. Support one or more Vulkan-capable ICD on a user's computer system without them interfering with one another.
  2. Support Vulkan Layers which are optional modules that can be enabled by an application, developer, or standard system settings.
  3. Impact the overall performance of a Vulkan application in the lowest possible fashion.

Layers

Layers are optional components that augment the Vulkan system. They can intercept, evaluate, and modify existing Vulkan functions on their way from the application down to the hardware. Layers are implemented as libraries that can be enabled in different ways (including by application request) and are loaded during CreateInstance. Each layer can choose to hook (intercept) any Vulkan functions which in turn can be ignored or augmented. A layer does not need to intercept all Vulkan functions. It may choose to intercept all known functions, or, it may choose to intercept only one function.

Some examples of features that layers may expose include:

  • Validating API usage
  • Adding the ability to perform Vulkan API tracing and debugging
  • Overlay additional content on the applications surfaces

Because layers are optionally, you may choose to enable layers for debugging your application, but then disable any layer usage when you release your product.

Installable Client Drivers

Vulkan allows multiple Installable Client Drivers (ICDs) each supporting one or more devices (represented by a Vulkan VkPhysicalDevice object) to be used collectively. The loader is responsible for discovering available Vulkan ICDs on the system. Given a list of available ICDs, the loader can enumerate all the physical devices available for an application and return this information to the application.

Instance Versus Device

There is an important concept which you will see brought up repeatedly throughout this document. Many functions, extensions, and other things in Vulkan are separated into two main groups:

  • Instance-related Objects
  • Device-related Objects
Instance-related Objects

A Vulkan Instance is a high-level construct used to provide Vulkan system-level information, or functionality. Vulkan objects associated directly with an instance are:

  • VkInstance
  • VkPhysicalDevice

An Instance function is any Vulkan function which takes as its first parameter either an object from the Instance list, or nothing at all. Some Vulkan Instance functions are:

  • vkEnumerateInstanceExtensionProperties
  • vkEnumeratePhysicalDevices
  • vkCreateInstance
  • vkDestroyInstance

You query Vulkan Instance functions using vkGetInstanceProcAddr. vkGetInstanceProcAddr can be used to query either device or instance entry- points in addition to all core entry-points. The returned function pointer is valid for this Instance and any object created under this Instance (including all VkDevice objects).

Similarly, an Instance extension is a set of Vulkan Instance functions extending the Vulkan language. These will be discussed in more detail later.

Device-related Objects

A Vulkan Device, on the other-hand, is a logical identifier used to associate functions with a particular physical device on a user's system. Vulkan constructs associated directly with a device include:

  • VkDevice
  • VkQueue
  • VkCommandBuffer
  • Any dispatchable object that is a child of a one of the above.

A Device function is any Vulkan function which takes any Device Object as its first parameter. Some Vulkan Device functions are:

  • vkQueueSubmit
  • vkBeginCommandBuffer
  • vkCreateEvent

You can query Vulkan Device functions using either vkGetInstanceProcAddr or vkGetDeviceProcAddr. If you choose to use vkGetInstanceProcAddr, it will have an additional level built into the call chain, which will reduce performance slightly. However, the function pointer returned can be used for any device created later, as long as it is associated with the same Vulkan Instance. If, instead you use vkGetDeviceProcAddr, the call chain will be more optimized to the specific device, but it will only work for the device used to query the function function pointer. Also, unlike vkGetInstanceProcAddr, vkGetDeviceProcAddr can only be used on core Vulkan Device functions, or Device extension functions.

The best solution is to query Instance extension functions using vkGetInstanceProcAddr, and to query Device extension functions using vkGetDeviceProcAddr. See Best Application Performance Setup for more information on this.

As with Instance extensions, a Device extension is a set of Vulkan Device functions extending the Vulkan language. You can read more about these later in the document.

Dispatch Tables and Call Chains

Vulkan uses an object model to control the scope of a particular action / operation. The object to be acted on is always the first parameter of a Vulkan call and is a dispatchable object (see Vulkan specification section 2.3 Object Model). Under the covers, the dispatchable object handle is a pointer to a structure, which in turn, contains a pointer to a dispatch table maintained by the loader. This dispatch table contains pointers to the Vulkan functions appropriate to that object.

There are two types of dispatch tables the loader maintains:

  • Instance Dispatch Table
  • Created in the loader during the call to vkCreateInstance
  • Device Dispatch Table
  • Created in the loader during the call to vkCreateDevice

At that time the application and/or system can specify optional layers to be included. The loader will initialize the specified layers to create a call chain for each Vulkan function and each entry of the dispatch table will point to the first element of that chain. Thus, the loader builds an instance call chain for each VkInstance that is created and a device call chain for each VkDevice that is created.

When an application calls a Vulkan function, this typically will first hit a trampoline function in the loader. These trampoline functions are small, simple functions that jump to the appropriate dispatch table entry for the object they are given. Additionally, for functions in the instance call chain, the loader has an additional function, called a terminator, which is called after all enabled layers to marshall the appropriate information to all available ICDs.

Instance Call Chain Example

For example, the diagram below represents what happens in the call chain for vkCreateInstance. After initializing the chain, the loader will call into the first layer's vkCreateInstance which will call the next finally terminating in the loader again where this function calls every ICD's vkCreateInstance and saves the results. This allows every enabled layer for this chain to set up what it needs based on the VkInstanceCreateInfo structure from the application.

Instance Call Chain

This also highlights some of the complexity the loader must manage when using instance call chains. As shown here, the loader's terminator must aggregate information to and from multiple ICDs when they are present. This implies that the loader has to be aware of any instance-level extensions which work on a VkInstance to aggregate them correctly.

Device Call Chain Example

Device call chains are created at vkCreateDevice and are generally simpler because they deal with only a single device and the ICD can always be the terminator of the chain.

Loader Device Call Chain



Application Interface to the Loader

In this section we'll discuss how an application interacts with the loader, including:

Interfacing with Vulkan Functions

There are several ways you can interface with Vulkan functions through the loader.

Vulkan Direct Exports

The loader library on Windows, Linux and Android will export all core Vulkan and all appropriate Window System Interface (WSI) extensions. This is done to make it simpler to get started with Vulkan development. When an application links directly to the loader library in this way, the Vulkan calls are simple trampoline functions that jump to the appropriate dispatch table entry for the object they are given.

Indirectly Linking to the Loader

Applications are not required to link directly to the loader library, instead they can use the appropriate platform specific dynamic symbol lookup on the loader library to initialize the application's own dispatch table. This allows an application to fail gracefully if the loader cannot be found. It also provides the fastest mechanism for the application to call Vulkan functions. An application will only need to query (via system calls such as dlsym()) the address of vkGetInstanceProcAddr from the loader library. Using vkGetInstanceProcAddr the application can then discover the address of all functions and extensions available, such as vkCreateInstance, vkEnumerateInstanceExtensionProperties and vkEnumerateInstanceLayerProperties in a platform-independent way.

Best Application Performance Setup

If you desire the best performance possible, you should setup your own dispatch table so that all your Instance functions are queried using vkGetInstanceProcAddr and all your Device functions are queried using vkGetDeviceProcAddr.

Why should you do this?

The answer comes in how the call chain of Instance functions are implemented versus the call chain of a Device functions. Remember, a [Vulkan Instance is a high-level construct used to provide Vulkan system-level information](#instance- related-objects). Because of this, Instance functions need to be broadcasted to every available ICD on the system. The following diagram shows an approximate view of an Instance call chain with 3 enabled layers:

Instance Call Chain

This is also how a Vulkan Device function call chain looks if you query it using vkGetInstanceProcAddr. On the otherhand, a Device function doesn't need to worry about the broadcast becuase it knows specifically which associated ICD and which associated Physical Device the call should terminate at. Because of this, the loader doesn't need to get involved between any enabled layers and the ICD. Thus, if you used a loader-exported Vulkan Device function, the call chain in the same scenario as above would look like:

Loader Device Call Chain

An even better solution would be for an application to perform a vkGetDeviceProcAddr call on all Device functions. This further optimizes the call chain by removing the loader all-together under most scenarios:

Application Device Call Chain

Also, notice if no layers are enabled, your application function pointer would point directly to the ICD. If called enough, those fewer calls can add up to performance savings.

NOTE: There are some Device functions which still require the loader to intercept them with a trampoline and terminator. There are very few of these, but they are typically functions which the loader wraps with its own data. In those cases, even the Device call chain will continue to look like the Instance call chain. One example of a Device function requiring a terminator is vkCreateSwapchainKHR. For that function, the loader needs to potentially convert the KHR_surface object into an ICD-specific KHR_surface object prior to passing down the rest of the function's information to the ICD.

Remember:

  • vkGetInstanceProcAddr can be used to query either device or instance entry-points in addition to all core entry-points.
  • vkGetDeviceProcAddr can only be used to query for device extension or core device entry-points.
ABI Versioning

The Vulkan loader library will be distributed in various ways including Vulkan SDKs, OS package distributions and Independent Hardware Vendor (IHV) driver packages. These details are beyond the scope of this document. However, the name and versioning of the Vulkan loader library is specified so an app can link to the correct Vulkan ABI library version. Vulkan versioning is such that ABI backwards compatibility is guaranteed for all versions with the same major number (e.g. 1.0 and 1.1). On Windows, the loader library encodes the ABI version in its name such that multiple ABI incompatible versions of the loader can peacefully coexist on a given system. The Vulkan loader library file name is vulkan-<ABI version>.dll. For example, for Vulkan version 1.X on Windows the library filename is vulkan-1.dll. And this library file can typically be found in the windows/system32 directory (on 64-bit Windows installs, the 32-bit version of the loader with the same name can be found in the windows/sysWOW64 directory).

For Linux, shared libraries are versioned based on a suffix. Thus, the ABI number is not encoded in the base of the library filename as on Windows. On Linux an application wanting to link to the latest Vulkan ABI version would just link to the name vulkan (libvulkan.so). A specific Vulkan ABI version can also be linked to by applications (e.g. libvulkan.so.1).

Application Layer Usage

Applications desiring Vulkan functionality beyond what the core API offers may use various layers or extensions. A layer cannot introduce new Vulkan core API entry-points to an application that are not exposed in Vulkan.h. However, layers may offer extensions that introduce new Vulkan commands that can be queried through the extension interface.

A common use of layers is for API validation which can be enabled by loading the layer during application development, but not loading the layer for application release. This eliminates the overhead of validating the application's usage of the API, something that wasn't available on some previous graphics APIs.

To find out what layers are available to your application, use vkEnumerateInstanceLayerProperties. This will report all layers that have been discovered by the loader. The loader looks in various locations to find layers on the system. For more information see the Layer discovery section below.

To enable a layer, or layers, simply pass the name of the layers you wish to enable in the ppEnabledLayerNames field of the VkInstanceCreateInfo during a call to vkCreateInstance. Once done, the layers you have enabled will be active for all Vulkan functions using the created VkInstance, and any of its child objects.

NOTE: Layer ordering is important in several cases since some layers interact with each other. Be careful when enabling layers as this may be the case. See the Overall Layer Ordering section for more information.

The following code section shows how you would go about enabling the VK_LAYER_LUNARG_standard_validation layer.

   char *instance_validation_layers[] = {
        "VK_LAYER_LUNARG_standard_validation"
    };
    const VkApplicationInfo app = {
        .sType = VK_STRUCTURE_TYPE_APPLICATION_INFO,
        .pNext = NULL,
        .pApplicationName = "TEST_APP",
        .applicationVersion = 0,
        .pEngineName = "TEST_ENGINE",
        .engineVersion = 0,
        .apiVersion = VK_API_VERSION_1_0,
    };
    VkInstanceCreateInfo inst_info = {
        .sType = VK_STRUCTURE_TYPE_INSTANCE_CREATE_INFO,
        .pNext = NULL,
        .pApplicationInfo = &app,
        .enabledLayerCount = 1,
        .ppEnabledLayerNames = (const char *const *)instance_validation_layers,
        .enabledExtensionCount = 0,
        .ppEnabledExtensionNames = NULL,
    };
    err = vkCreateInstance(&inst_info, NULL, &demo->inst);

At vkCreateInstance and vkCreateDevice, the loader constructs call chains that include the application specified (enabled) layers. Order is important in the ppEnabledLayerNames array; array element 0 is the topmost (closest to the application) layer inserted in the chain and the last array element is closest to the driver. See the Overall Layer Ordering section for more information on layer ordering.

NOTE: Device Layers Are Now Deprecated

vkCreateDevice originally was able to select layers in a similar manner to vkCreateInstance. This lead to the concept of "instance layers" and "device layers". It was decided by Khronos to deprecate the "device layer" functionality and only consider "instance layers". Therefore, vkCreateDevice will use the layers specified at vkCreateInstance. Because of this, the following items have been deprecated:

  • VkDeviceCreateInfo fields:
  • ppEnabledLayerNames
  • enabledLayerCount
  • The vkEnumerateDeviceLayerProperties function
Implicit vs Explicit Layers

Explicit layers are layers which are enabled by an application (e.g. with the vkCreateInstance function), or by an environment variable (as mentioned previously).

Implicit layers are those which are enabled by their existence. For example, certain application environments (e.g. Steam or an automotive infotainment system) may have layers which they always want enabled for all applications that they start. Other implicit layers may be for all applications started on a given system (e.g. layers that overlay frames-per-second). Implicit layers are enabled automatically, whereas explicit layers must be enabled explicitly.

Implicit layers have an additional requirement over explicit layers in that they require being able to be disabled by an environmental variable. This is due to the fact that they are not visible to the application and could cause issues. A good principle to keep in mind would be to define both an enable and disable environment variable so the users can deterministicly enable the functionality. On Desktop platforms (Windows and Linux), these enable/disable settings are defined in the layer's JSON file.

Discovery of system-installed implicit and explicit layers is described later in the Layer Discovery Section. For now, simply know that what distinguishes a layer as implicit or explicit is dependent on the Operating system, as shown in the table below.

Operating System Implicit Layer Identification
Windows Implicit Layers are located in a different Windows registry location than Explicit Layers.
Linux Implicit Layers are located in a different directory location than Explicit Layers.
Android There is No Support For Implicit Layers on Android.
Forcing Layer Source Folders

Developers may need to use special, pre-production layers, without modifying the system-installed layers. You can direct the loader to look for layers in a specific folder by defining the "VK_LAYER_PATH" environment variable. This will override the mechanism used for finding system-installed layers. Because layers of interest may exist in several disinct folders on a system, this environment variable can containis several paths seperated by the operating specific path separator. On Windows, each separate folder should be separated in the list using a semi-colon. On Linux, each folder name should be separated using a colon.

If "VK_LAYER_PATH" exists, only the folders listed in it will be scanned for layers. Each directory listed should be the full pathname of a folder containing layer manifest files.

Forcing Layers to be Enabled on Windows and Linux

Developers may want to enable layers that are not enabled by the given application they are using. On Linux and Windows, the environment variable "VK_INSTANCE_LAYERS" can be used to enable additional layers which are not specified (enabled) by the application at vkCreateInstance. "VK_INSTANCE_LAYERS" is a colon (Linux)/semi-colon (Windows) separated list of layer names to enable. Order is relevant with the first layer in the list being the top-most layer (closest to the application) and the last layer in the list being the bottom-most layer (closest to the driver). See the Overall Layer Ordering section for more information.

Application specified layers and user specified layers (via environment variables) are aggregated and duplicates removed by the loader when enabling layers. Layers specified via environment variable are top-most (closest to the application) while layers specified by the application are bottommost.

An example of using these environment variables to activate the validation layer VK_LAYER_LUNARG_parameter_validation on Windows or Linux is as follows:

> $ export VK_INSTANCE_LAYERS=VK_LAYER_LUNARG_parameter_validation
Overall Layer Ordering

The overall ordering of all layers by the loader based on the above looks as follows:

Loader Layer Ordering

Ordering may also be important internal to the list of Explicit Layers. Some layers may be dependent on other behavior being implemented before or after the loader calls it. For example: the VK_LAYER_LUNARG_core_validation layer expects the VK_LAYER_LUNARG_parameter_validation to be called first. This is because the VK_LAYER_LUNARG_parameter_validation will filter out any invalid NULL pointer calls prior to the rest of the validation checking done by VK_LAYER_LUNARG_core_validation. If not done properly, you may see crashes in the VK_LAYER_LUNARG_core_validation layer that would otherwise be avoided.

Application Usage of Extensions

Extensions are optional functionality provided by a layer, the loader or an ICD. Extensions can modify the behavior of the Vulkan API and need to be specified and registered with Khronos. These extensions can be created by an Independent Hardware Vendor (IHV) to expose new hardware functionality, or by a layer writer to expose some internal feature, or by the loader to improve functional behavior. Information about various extensions can be found in the Vulkan Spec, and vulkan.h header file.

Instance and Device Extensions

As hinted at in the Instance Versus Device section, there are really two types of extensions:

  • Instance Extensions
  • Device Extensions

An Instance extension is an extension which modifies existing behavior or implements new behavior on instance-level objects, like a VkInstance or a VkPhysicalDevice. A Device extension is an extension which does the same, but for any VkDevice object, or any dispatchable object that is a child of a VkDevice (VkQueue and VkCommandBuffer are examples of these).

It is very important to know what type of extension you are desiring to enable as you will enable Instance extensions during vkCreateInstance and Device extensions during vkCreateDevice.

The loader discovers and aggregates all extensions from layers (both explicit and implicit), ICDs and the loader before reporting them to the application in vkEnumerateXXXExtensionProperties (where XXX is either "Instance" or "Device").

  • Instance extensions are discovered via vkEnumerateInstanceExtensionProperties.
  • Device extensions are be discovered via vkEnumerateDeviceExtensionProperties.

Looking at vulkan.h, you'll notice that they are both similar. For example, vkEnumerateInstanceExtensionProperties prototype looks as follows:

   VkResult
   vkEnumerateInstanceExtensionProperties(const char *pLayerName,
                                          uint32_t *pPropertyCount,
                                          VkExtensionProperties *pProperties);

The "pLayerName" parameter in these functions is used to select either a single layer or the Vulkan platform implementation. If "pLayerName" is NULL, extensions from Vulkan implementation components (including loader, implicit layers, and ICDs) are enumerated. If "pLayerName" is equal to a discovered layer module name then only extensions from that layer (which may be implicit or explicit) are enumerated. Duplicate extensions (e.g. an implicit layer and ICD might report support for the same extension) are eliminated by the loader. For duplicates, the ICD version is reported and the layer version is culled.

Also, Extensions must be enabled (in vkCreateInstance or vkCreateDevice) before the functions associated with the extensions can be used. If you get an Extension function using either vkGetInstanceProcAddr or vkGetDeviceProcAddr, but fail to enable it, you could experience undefined behavior. This should actually be flagged if you run with Validation layers enabled.

WSI Extensions

Khronos approved WSI extensions are available and provide Windows System Integration support for various execution environments. It is important to understand that some WSI extensions are valid for all targets, but others are particular to a given execution environment (and loader). This desktop loader (currently targeting Windows and Linux) only enables and directly exports those WSI extensions that are appropriate to the current environment. For the most part, the selection is done in the loader using compile-time preprocessor flags. All versions of the desktop loader currently expose at least the following WSI extension support:

  • VK_KHR_surface
  • VK_KHR_swapchain
  • VK_KHR_display

In addition, each of the following OS targets for the loader support target- specific extensions:

Windowing System Extensions available
Windows VK_KHR_win32_surface
Linux (Default) VK_KHR_xcb_surface and VK_KHR_xlib_surface
Linux (Wayland) VK_KHR_wayland_surface
Linux (Mir) VK_KHR_mir_surface

NOTE: Wayland and Mir targets are not fully supported at this time. Wayland support is present, but should be considered Beta quality. Mir support is not completely implemented at this time.

It is important to understand that while the loader may support the various entry-points for these extensions, there is a hand-shake required to actually use them:

  • At least one physical device must support the extension(s)
  • The application must select such a physical device
  • The application must request the extension(s) be enabled while creating the instance or logical device (This depends on whether or not the given extension works with an instance or a device).
  • The instance and/or logical device creation must succeed.

Only then can you expect to properly use a WSI extension in your Vulkan program.

Unknown Extensions

With the ability to expand Vulkan so easily, extensions will be created that the loader knows nothing about. If the extension is a device extension, the loader will pass the unknown entry-point down the device call chain ending with the appropriate ICD entry-points. The same thing will happen, if the extension is an instance extension which takes a physical device paramater as it's first component. However, for all other instance extensions the loader will fail to load it.

But why doesn't the loader support unknown instance extensions?
Let's look again at the Instance call chain:

Instance call chain

Notice that for a normal instance function call, the loader has to handle passing along the function call to the available ICDs. If the loader has no idea of the parameters or return value of the instance call, it can't properly pass information along to the ICDs. There may be ways to do this, which will be explored in the future. However, for now, this loader does not support instance extensions which don't take a physical device as their first parameter.

Because the device call-chain does not normally pass through the loader terminator, this is not a problem for device extensions. Additionally, since a physical device is associated with one ICD, we can use a generic terminator pointing to one ICD. This is because both of these extensions terminate directly in the ICD they are associated with.

Is this a big problem?
No! Most extension functionality only affects either a physical or logical device and not an instance. Thus, the overwhelming majority of extensions should be supported with direct loader support.

Filtering Out Unknown Instance Extension Names

In some cases, an ICD may support instance extensions that the loader does not. For the above reasons, the loader will filter out the names of these unknown instance extensions when an application calls vkEnumerateInstanceExtensionProperties. Additionally, this behavior will cause the loader to throw an error during vkCreateInstance if you still attempt to use one of these extensions. The intent is to protect applications so that they don't inadvertantly use functionality which could lead to a crash.

On the other-hand, if you know you can safely use the extension, you may disable the filtering by defining the environment variable VK_LOADER_DISABLE_INST_EXT_FILTER and setting the value to a non-zero number. This will effectively disable the loader's filtering out of instance extension names.



Loader and Layer Interface

In this section we'll discuss how the loader interacts with layers, including:

Layer Discovery

As mentioned in the Application Interface section, layers can be categorized into two categories:

  • Implicit Layers
  • Explicit Layers

The main difference between the two is that Implicit Layers are automatically enabled, unless overriden, and Explicit Layers must be enabled. Remember, Implicit Layers are not present on all Operating Systems (like Android).

On any system, the loader looks in specific areas for information on the layers that it can load at a user's request. The process of finding the available layers on a system is known as Layer Discovery. During discovery, the loader determines what layers are available, the layer name, the layer version, and any extensions supported by the layer. This information is provided back to an application through vkEnumerateInstanceLayerProperties.

The group of layers available to the loader is known as a layer library. This section defines an extensible interface to discover what layers are contained in the layer library.

This section also specifies the minimal conventions and rules a layer must follow, especially with regards to interacting with the loader and other layers.

Layer Manifest File Usage

On Windows and Linux systems, JSON formatted manifest files are used to store layer information. In order to find system-installed layers, the Vulkan loader will read the JSON files to identify the names and attributes of layers and their extensions. The use of manifest files allows the loader to avoid loading any shared library files when the application does not query nor request any extensions. The format of Layer Manifest File is detailed below.

The Android loader does not use manifest files. Instead, the loader queries the layer properties using special functions known as "introspection" functions. The intent of these functions is to determine the same required information gathered from reading the manifest files. These introspection functions are not used by the desktop loader but should be present in layers to maintain consistency. The specific "introspection" functions are called out in the Layer Manifest File Format table.

Android Layer Discovery

On Android, the loader looks for layers to enumerate in the /data/local/debug/vulkan folder. An application enabled for debug has the ability to enumerate and enable any layers in that location.

Windows Layer Discovery

In order to find system-installed layers, the Vulkan loader will scan the values in the following Windows registry keys:

   HKEY_LOCAL_MACHINE\SOFTWARE\Khronos\Vulkan\ExplicitLayers
   HKEY_CURRENT_USER\SOFTWARE\Khronos\Vulkan\ExplicitLayers
   HKEY_LOCAL_MACHINE\SOFTWARE\Khronos\Vulkan\ImplicitLayers
   HKEY_CURRENT_USER\SOFTWARE\Khronos\Vulkan\ImplicitLayers

For each value in these keys which has DWORD data set to 0, the loader opens the JSON manifest file specified by the name of the value. Each name must be a full pathname to the manifest file. The Vulkan loader will open each info file to obtain information about the layer, including the name or pathname of a shared library (".dll") file. However, if VK_LAYER_PATH is defined, then the loader will instead look at the paths defined by that variable instead of using the information provided by these registry keys. See Forcing Layer Source Folders for more information on this.

Linux Layer Discovery

On Linux, the Vulkan loader will scan the files in the following Linux directories:

/usr/local/etc/vulkan/explicit_layer.d
/usr/local/etc/vulkan/implicit_layer.d
/usr/local/share/vulkan/explicit_layer.d
/usr/local/share/vulkan/implicit_layer.d
/etc/vulkan/explicit_layer.d
/etc/vulkan/implicit_layer.d
/usr/share/vulkan/explicit_layer.d
/usr/share/vulkan/implicit_layer.d
$HOME/.local/share/vulkan/explicit_layer.d
$HOME/.local/share/vulkan/implicit_layer.d

Of course, ther are some things you have to know about the above folders:

  1. The "/usr/local/*" directories can be configured to be other directories at build time.
  2. $HOME is the current home directory of the application's user id; this path will be ignored for suid programs.
  3. The "/usr/local/etc/vulkan/*_layer.d" and "/usr/local/share/vulkan/*_layer.d" directories are for layers that are installed from locally-built sources.
  4. The "/usr/share/vulkan/*_layer.d" directories are for layers that are installed from Linux-distribution-provided packages.

As on Windows, if VK_LAYER_PATH is defined, then the loader will instead look at the paths defined by that variable instead of using the information provided by these default paths. However, these environment variables are only used for non-suid programs. See Forcing Layer Source Folders for more information on this.

Layer Version Negotiation

Now that a layer has been discovered, an application can choose to load it (or it is loaded by default if it is an Implicit layer). When the loader attempts to load the layer, the first thing it does is attempt to negotiate the version of the loader to layer interface. In order to negotiate the loader/layer interface version, the layer must implement the vkNegotiateLoaderLayerInterfaceVersion function. The following information is provided for this interface in include/vulkan/vk_layer.h:

  typedef enum VkNegotiateLayerStructType {
      LAYER_NEGOTIATE_INTERFACE_STRUCT = 1,
  } VkNegotiateLayerStructType;

  typedef struct VkNegotiateLayerInterface {
      VkNegotiateLayerStructType sType;
      void *pNext;
      uint32_t loaderLayerInterfaceVersion;
      PFN_vkGetInstanceProcAddr pfnGetInstanceProcAddr;
      PFN_vkGetDeviceProcAddr pfnGetDeviceProcAddr;
      PFN_GetPhysicalDeviceProcAddr pfnGetPhysicalDeviceProcAddr;
  } VkNegotiateLayerInterface;

  VkResult vkNegotiateLoaderLayerInterfaceVersion(
                   VkNegotiateLayerInterface *pVersionStruct);

You'll notice the VkNegotiateLayerInterface structure is similar to other Vulkan structures. The "sType" field, in this case takes a new enum defined just for internal loader/layer interfacing use. The valid values for "sType" could grow in the future, but right only havs the one value "LAYER_NEGOTIATE_INTERFACE_STRUCT".

This function (vkNegotiateLoaderLayerInterfaceVersion) should be exported by the layer so that using "GetProcAddress" on Windows or "dlsym" on Linux, should return a valid function pointer to it. Once the loader has grabbed a valid address to the layers function, the loader will create a variable of type VkNegotiateLayerInterface and initialize it in the following ways:

  1. Set the structure "sType" to "LAYER_NEGOTIATE_INTERFACE_STRUCT"
  2. Set pNext to NULL.
    • This is for future growth
  3. Set "loaderLayerInterfaceVersion" to the current version the loader desires to set the interface to.
    • The minimum value sent by the loader will be 2 since it is the first version supporting this function.

The loader will then individually call each layer’s vkNegotiateLoaderLayerInterfaceVersion function with the filled out “VkNegotiateLayerInterface”. The layer will either accept the loader's version set in "loaderLayerInterfaceVersion", or modify it to the closest value version of the interface that the layer can support. The value should not be higher than the version requested by the loader. If the layer can't support at a minimum the version requested, then the layer should return an error like "VK_ERROR_INITIALIZATION_FAILED". If a layer can support some version, then the layer should do the following:

  1. Adjust the version to the layer's desired version.
  2. The layer should fill in the function pointer values to its internal functions:
    • "pfnGetInstanceProcAddr" should be set to the layer’s internal GetInstanceProcAddr function.
    • "pfnGetDeviceProcAddr" should be set to the layer’s internal GetDeviceProcAddr function.
    • "pfnGetPhysicalDeviceProcAddr" should be set to the layer’s internal GetPhysicalDeviceProcAddr function.
      • If the layer supports no physical device extensions, it may set the value to NULL.
      • More on this function later
  3. The layer should return "VK_SUCCESS"

This function SHOULD NOT CALL DOWN the layer chain to the next layer. The loader will work with each layer individually.

If the layer supports the new interface and reports version 2 or greater, then the loader will use the “fpGetInstanceProcAddr” and “fpGetDeviceProcAddr” functions from the “VkNegotiateLayerInterface” structure. Prior to these changes, the loader would query each of those functions using "GetProcAddress" on Windows or "dlsym" on Linux.

Layer Call Chains and Distributed Dispatch

There are two key architectural features that drive the loader to layer library interface:

  1. Separate and distinct instance and device call chains
  2. Distributed dispatch.

You can read an overview of dispatch tables and call chains above in the Dispatch Tables and Call Chains section.

What's important to note here is that a layer can intercept Vulkan instance functions, device functions or both. For a layer to intercept instance functions, it must participate in the instance call chain. For a layer to intercept device functions, it must participate in the device call chain.

Remember, a layer does not need to intercept all instance or device functions, instead, it can choose to intercept only a subset of those functions.

Normally, when a layer intercepts a given Vulkan function, it will call down the instance or device call chain as needed. The loader and all layer libraries that participate in a call chain cooperate to ensure the correct sequencing of calls from one entity to the next. This group effort for call chain sequencing is hereinafter referred to as distributed dispatch.

In distributed dispatch each layer is responsible for properly calling the next entity in the call chain. This means that a dispatch mechanism is required for all Vulkan functions that a layer intercepts. If a Vulkan function is not intercepted by a layer, or if a layer chooses to terminate the function by not calling down the chain, then no dispatch is needed for that particular function.

For example, if the enabled layers intercepted only certain instance functions, the call chain would look as follows: Instance Function Chain

Likewise, if the enabled layers intercepted only a few of the device functions, the call chain could look this way: Device Function Chain

The loader is responsible for dispatching all core and instance extension Vulkan functions to the first entity in the call chain.

Layer Unknown Physical Device Extensions

Originally, if the loader was called with vkGetInstanceProcAddr, it would result in the following behavior:

  1. The loader would check if core function:
    • If it was, it would return the function pointer
  2. The loader would check if known extension function:
    • If it was, it would return the function pointer
  3. If the loader knew nothing about it, it would call down using GetInstanceProcAddr
    • If it returned non-NULL, treat it as an unknown logical device command.
    • This meant setting up a generic trampoline function that takes in a VkDevice as the first parameter and adjusting the dispatch table to call the ICD/Layers function after getting the dispatch table from the VkDevice.
  4. If all the above failed, the loader would return NULL to the application.

This caused problems when a layer attempted to expose new physical device extensions the loader knew nothing about, but an application did. Because the loader knew nothing about it, the loader would get to step 3 in the above process and would treat the function as an unknown logical device command. The problem is, this would create a generic VkDevice trampoline function which, on the first call, would attempt to dereference the VkPhysicalDevice as a VkDevice. This would lead to a crash or corruption.

In order to identify the extension entry-points specific to physical device extensions, the following function can be added to a layer:

PFN_vkVoidFunction vk_layerGetPhysicalDeviceProcAddr(VkInstance instance,
                                                     const char* pName);

This function behaves similar to vkGetInstanceProcAddr and vkGetDeviceProcAddr except it should only return values for physical device extension entry-points. In this way, it compares "pName" to every physical device function supported in the layer.

The following rules apply:

  • If it is the name of a physical device function supported by the layer, the pointer to the layer's corresponding function should be returned.
  • If it is the name of a valid function which is not a physical device function (i.e. an Instance, Device, or other function implemented by the layer), then the value of NULL should be returned.
    • We don’t call down since we know the command is not a physical device extension).
  • If the layer has no idea what this function is, it should call down the layer chain to the next vk_layerGetPhysicalDeviceProcAddr call.
    • This can be retrieved in one of two ways:
      • During vkCreateInstance, it is passed to a layer in the chain information passed to a layer in the VkLayerInstanceCreateInfo structure.
        • Use get_chain_info() to get the pointer to the VkLayerInstanceCreateInfo structure. Let's call it chain_info.
        • The address is then under chain_info->u.pLayerInfo->pfnNextGetPhysicalDeviceProcAddr
        • See Example Code for CreateInstance
      • Using the next layer’s GetInstanceProcAddr function to query for vk_layerGetPhysicalDeviceProcAddr.

This support is optional and should not be considered a requirement. This is only required if a layer intends to support some functionality not directly supported by loaders released in the public. If a layer does implement this support, it should return the address of its vk_layerGetPhysicalDeviceProcAddr function in the "pfnGetPhysicalDeviceProcAddr" member of the VkNegotiateLayerInterface structure during Layer Version Negotiation. Additionally, the layer should also make sure vkGetInstanceProcAddr returns a valid function pointer to a query of vk_layerGetPhysicalDeviceProcAddr.

The new behavior of the loader's vkGetInstanceProcAddr with support for the vk_layerGetPhysicalDeviceProcAddr function is as follows:

  1. Check if core function:
    • If it is, return the function pointer
  2. Check if known instance or device extension function:
    • If it is, return the function pointer
  3. Call the layer/ICD GetPhysicalDeviceProcAddr
    • If it returns non-NULL, return a trampoline to a generic physical device function, and setup a generic terminator which will pass it to the proper ICD.
  4. Call down using GetInstanceProcAddr
    • If it returns non-NULL, treat it as an unknown logical device command. This means setting up a generic trampoline function that takes in a VkDevice as the first parameter and adjusting the dispatch table to call the ICD/Layers function after getting the dispatch table from the VkDevice. Then, return the pointer to corresponding trampoline function.
  5. Return NULL

You can see now, that, if the command gets promoted to core later, it will no longer be setup using vk_layerGetPhysicalDeviceProcAddr. Additionally, if the loader adds direct support for the extension, it will no longer get to step 3, because step 2 will return a valid function pointer. However, the layer should continue to support the command query via vk_layerGetPhysicalDeviceProcAddr, until at least a Vulkan version bump, because an older loader may still be attempting to use the commands.

Layer Intercept Requirements

  • Layers intercept a Vulkan function by defining a C/C++ function with signature identical to the Vulkan API for that function.
  • A layer must intercept at least vkGetInstanceProcAddr and vkCreateInstance to participate in the instance call chain.
  • A layer may also intercept vkGetDeviceProcAddr and vkCreateDevice to participate in the device call chain.
  • For any Vulkan function a layer intercepts which has a non-void return value, an appropriate value must be returned by the layer intercept function.
  • Most functions a layer intercepts should call down the chain to the corresponding Vulkan function in the next entity.
  • The common behavior for a layer is to intercept a call, perform some behavior, then pass it down to the next entity.
    • If you don't pass the information down, undefined behavior may occur.
    • This is because the function will not be received by layers further down the chain, or any ICDs.
  • One function that must never call down the chain is:
    • vkNegotiateLoaderLayerInterfaceVersion
  • Three common functions that may not call down the chain are:
    • vkGetInstanceProcAddr
    • vkGetDeviceProcAddr
    • vk_layerGetPhysicalDeviceProcAddr
    • These functions only call down the chain for Vulkan functions that they do not intercept.
  • Layer intercept functions may insert extra calls to Vulkan functions in addition to the intercept.
  • For example, a layer intercepting vkQueueSubmit may want to add a call to vkQueueWaitIdle after calling down the chain for vkQueueSubmit.
  • This would result in two calls down the chain: First a call down the vkQueueSubmit chain, followed by a call down the vkQueueWaitIdle chain.
  • Any additional calls inserted by a layer must be on the same chain
    • If the function is a device function, only other device functions should be added.
    • Likewise, if the function is an instance function, only other instance functions should be added.

Distributed Dispatching Requirements

  • For each entry-point a layer intercepts, it must keep track of the entry point residing in the next entity in the chain it will call down into.
  • In other words, the layer must have a list of pointers to functions of the appropriate type to call into the next entity.
  • This can be implemented in various ways but for clarity, will be referred to as a dispatch table.
  • A layer can use the VkLayerDispatchTable structure as a device dispatch table (see include/vulkan/vk_layer.h).
  • A layer can use the VkLayerInstanceDispatchTable structure as a instance dispatch table (see include/vulkan/vk_layer.h).
  • A Layer's vkGetInstanceProcAddr function uses the next entity's vkGetInstanceProcAddr to call down the chain for unknown (i.e. non-intercepted) functions.
  • A Layer's vkGetDeviceProcAddr function uses the next entity's vkGetDeviceProcAddr to call down the chain for unknown (i.e. non-intercepted) functions.
  • A Layer's vk_layerGetPhysicalDeviceProcAddr function uses the next entity's vk_layerGetPhysicalDeviceProcAddr to call down the chain for unknown (i.e. non-intercepted) functions.

Layer Conventions and Rules

A layer, when inserted into an otherwise compliant Vulkan implementation, must still result in a compliant Vulkan implementation. The intention is for layers to have a well-defined baseline behavior. Therefore, it must follow some conventions and rules defined below.

A layer is always chained with other layers. It must not make invalid calls to, or rely on undefined behaviors of, its lower layers. When it changes the behavior of a function, it must make sure its upper layers do not make invalid calls to or rely on undefined behaviors of its lower layers because of the changed behavior. For example, when a layer intercepts an object creation function to wrap the objects created by its lower layers, it must make sure its lower layers never see the wrapping objects, directly from itself or indirectly from its upper layers.

When a layer requires host memory, it may ignore the provided allocators. It should use memory allocators if the layer is intended to run in a production environment. For example, this usually applies to implicit layers that are always enabled. That will allow applications to include the layer's memory usage.

Additional rules include:

  • vkEnumerateInstanceLayerProperties must enumerate and only enumerate the layer itself.
  • vkEnumerateInstanceExtensionProperties must handle the case where pLayerName is itself.
  • It must return VK_ERROR_LAYER_NOT_PRESENT otherwise, including when pLayerName is NULL.
  • vkEnumerateDeviceLayerProperties is deprecated and may be omitted.
  • Using this will result in undefined behavior.
  • vkEnumerateDeviceExtensionProperties must handle the case where pLayerName is itself.
  • In other cases, it should normally chain to other layers.
  • vkCreateInstance must not generate an error for unrecognized layer names and extension names.
  • It may assume the layer names and extension names have been validated.
  • vkGetInstanceProcAddr intercepts a Vulkan function by returning a local entry-point
  • Otherwise it returns the value obtained by calling down the instance call chain.
  • vkGetDeviceProcAddr intercepts a Vulkan function by returning a local entry-point
  • Otherwise it returns the value obtained by calling down the device call chain.
  • These additional functions must be intercepted if the layer implements device-level call chaining:
  • vkGetDeviceProcAddr
  • vkCreateDevice(only required for any device-level chaining)
    • NOTE: older layer libraries may expect that vkGetInstanceProcAddr ignore instance when pName is vkCreateDevice.
  • The specification requires NULL to be returned from vkGetInstanceProcAddr and vkGetDeviceProcAddr for disabled functions.
  • A layer may return NULL itself or rely on the following layers to do so.

Layer Dispatch Initialization

  • A layer initializes its instance dispatch table within its vkCreateInstance function.
  • A layer initializes its device dispatch table within its vkCreateDevice function.
  • The loader passes a linked list of initialization structures to layers via the "pNext" field in the VkInstanceCreateInfo and VkDeviceCreateInfo structures for vkCreateInstance and VkCreateDevice respectively.
  • The head node in this linked list is of type VkLayerInstanceCreateInfo for instance and VkLayerDeviceCreateInfo for device. See file include/vulkan/vk_layer.h for details.
  • A VK_STRUCTURE_TYPE_LOADER_INSTANCE_CREATE_INFO is used by the loader for the "sType" field in VkLayerInstanceCreateInfo.
  • A VK_STRUCTURE_TYPE_LOADER_DEVICE_CREATE_INFO is used by the loader for the "sType" field in VkLayerDeviceCreateInfo.
  • The "function" field indicates how the union field "u" should be interpreted within VkLayer*CreateInfo. The loader will set the "function" field to VK_LAYER_LINK_INFO. This indicates "u" field should be VkLayerInstanceLink or VkLayerDeviceLink.
  • The VkLayerInstanceLink and VkLayerDeviceLink structures are the list nodes.
  • The VkLayerInstanceLink contains the next entity's vkGetInstanceProcAddr used by a layer.
  • The VkLayerDeviceLink contains the next entity's vkGetInstanceProcAddr and vkGetDeviceProcAddr used by a layer.
  • Given the above structures set up by the loader, layer must initialize their dispatch table as follows:
    • Find the VkLayerInstanceCreateInfo/VkLayerDeviceCreateInfo structure in the VkInstanceCreateInfo/VkDeviceCreateInfo structure.
    • Get the next entity's vkGet*ProcAddr from the "pLayerInfo" field.
    • For CreateInstance get the next entity's vkCreateInstance by calling the "pfnNextGetInstanceProcAddr": pfnNextGetInstanceProcAddr(NULL, "vkCreateInstance").
    • For CreateDevice get the next entity's vkCreateDevice by calling the "pfnNextGetInstanceProcAddr": pfnNextGetInstanceProcAddr(NULL, "vkCreateDevice").
    • Advanced the linked list to the next node: pLayerInfo = pLayerInfo->pNext.
    • Call down the chain either vkCreateDevice or vkCreateInstance
    • Initialize your layer dispatch table by calling the next entity's Get*ProcAddr function once for each Vulkan function needed in your dispatch table

Example Code for CreateInstance

VkResult vkCreateInstance(
        const VkInstanceCreateInfo *pCreateInfo,
        const VkAllocationCallbacks *pAllocator,
        VkInstance *pInstance)
{
   VkLayerInstanceCreateInfo *chain_info =
        get_chain_info(pCreateInfo, VK_LAYER_LINK_INFO);

    assert(chain_info->u.pLayerInfo);
    PFN_vkGetInstanceProcAddr fpGetInstanceProcAddr =
        chain_info->u.pLayerInfo->pfnNextGetInstanceProcAddr;
    PFN_vkCreateInstance fpCreateInstance =
        (PFN_vkCreateInstance)fpGetInstanceProcAddr(NULL, "vkCreateInstance");
    if (fpCreateInstance == NULL) {
        return VK_ERROR_INITIALIZATION_FAILED;
    }

    // Advance the link info for the next element of the chain
    chain_info->u.pLayerInfo = chain_info->u.pLayerInfo->pNext;

    // Continue call down the chain
    VkResult result = fpCreateInstance(pCreateInfo, pAllocator, pInstance);
    if (result != VK_SUCCESS)
        return result;

    // Init layer's dispatch table using GetInstanceProcAddr of
    // next layer in the chain.
    instance_dispatch_table = new VkLayerInstanceDispatchTable;
    layer_init_instance_dispatch_table(
        *pInstance, my_data->instance_dispatch_table, fpGetInstanceProcAddr);

    // Other layer initialization
    ...

    return VK_SUCCESS;
}

Example Code for CreateDevice

VkResult 
vkCreateDevice(
        VkPhysicalDevice gpu,
        const VkDeviceCreateInfo *pCreateInfo,
        const VkAllocationCallbacks *pAllocator,
        VkDevice *pDevice)
{
    VkLayerDeviceCreateInfo *chain_info =
        get_chain_info(pCreateInfo, VK_LAYER_LINK_INFO);

    PFN_vkGetInstanceProcAddr fpGetInstanceProcAddr =
        chain_info->u.pLayerInfo->pfnNextGetInstanceProcAddr;
    PFN_vkGetDeviceProcAddr fpGetDeviceProcAddr =
        chain_info->u.pLayerInfo->pfnNextGetDeviceProcAddr;
    PFN_vkCreateDevice fpCreateDevice =
        (PFN_vkCreateDevice)fpGetInstanceProcAddr(NULL, "vkCreateDevice");
    if (fpCreateDevice == NULL) {
        return VK_ERROR_INITIALIZATION_FAILED;
    }

    // Advance the link info for the next element on the chain
    chain_info->u.pLayerInfo = chain_info->u.pLayerInfo->pNext;

    VkResult result = fpCreateDevice(gpu, pCreateInfo, pAllocator, pDevice);
    if (result != VK_SUCCESS) {
        return result;
    }

    // initialize layer's dispatch table
    device_dispatch_table = new VkLayerDispatchTable;
    layer_init_device_dispatch_table(
        *pDevice, device_dispatch_table, fpGetDeviceProcAddr);

    // Other layer initialization
    ...

    return VK_SUCCESS;
}

Special Considerations

Associating Private Data with Vulkan Objects Within a Layer

A layer may want to associate it's own private data with one or more Vulkan objects. Two common methods to do this are hash maps and object wrapping.

Wrapping

The loader supports layers wrapping any Vulkan object, including dispatchable objects. For functions that return object handles, each layer does not touch the value passed down the call chain. This is because lower items may need to use the original value. However, when the value is returned from a lower-level layer (possibly the ICD), the layer saves the handle and returns its own handle to the layer above it (possibly the application). When a layer receives a Vulkan function using something that it previously returned a handle for, the layer is required to unwrap the handle and pass along the saved handle to the layer below it. This means that the layer must intercept every Vulkan function which uses the object in question, and wrap or unwrap the object, as appropriate. This includes adding support for all extensions with functions using any object the layer wraps.

Layers above the object wrapping layer will see the wrapped object. Layers which wrap dispatchable objects must ensure that the first field in the wrapping structure is a pointer to a dispatch table as defined in vk_layer.h. Specifically, an instance wrapped dispatchable object could be as follows:

struct my_wrapped_instance_obj_ {
    VkLayerInstanceDispatchTable *disp;
    // whatever data layer wants to add to this object
};

A device wrapped dispatchable object could be as follows:

struct my_wrapped_instance_obj_ {
    VkLayerDispatchTable *disp;
    // whatever data layer wants to add to this object
};

Layers that wrap dispatchable objects must follow the guidelines for creating new dispatchable objects (below).

Cautions About Wrapping

Layers are generally discouraged from wrapping objects, because of the potential for incompatibilities with new extensions. For example, let's say that a layer wraps VkImage objects, and properly wraps and unwraps VkImage object handles for all core functions. If a new extension is created which has functions that take VkImage objects as parameters, and if the layer does not support those new functions, an application that uses both the layer and the new extension will have undefined behavior when those new functions are called (e.g. the application may crash). This is because the lower-level layers and ICD won't receive the handle that they generated. Instead, they will receive a handle that is only known by the layer that is wrapping the object.

Because of the potential for incompatibilities with unsupported extensions, layers that wrap objects must check which extensions are being used by the application, and take appropriate action if the layer is used with unsupported extensions (e.g. disable layer functionality, stop wrapping objects, issue a message to the user).

The reason that the validation layers wrap objects, is to track the proper use and destruction of each object. They issue a validation error if used with unsupported extensions, alerting the user to the potential for undefined behavior.

Hash Maps

Alternatively, a layer may want to use a hash map to associate data with a given object. The key to the map could be the object. Alternatively, for dispatchable objects at a given level (eg device or instance) the layer may want data associated with the VkDevice or VkInstance objects. Since there are multiple dispatchable objects for a given VkInstance or VkDevice, the VkDevice or VkInstance object is not a great map key. Instead the layer should use the dispatch table pointer within the VkDevice or VkInstance since that will be unique for a given VkInstance or VkDevice.

Creating New Dispatchable Objects

Layers which create dispatchable objects must take special care. Remember that loader trampoline code normally fills in the dispatch table pointer in the newly created object. Thus, the layer must fill in the dispatch table pointer if the loader trampoline will not do so. Common cases where a layer (or ICD) may create a dispatchable object without loader trampoline code is as follows:

  • layers that wrap dispatchable objects
  • layers which add extensions that create dispatchable objects
  • layers which insert extra Vulkan functions in the stream of functions they intercept from the application
  • ICDs which add extensions that create dispatchable objects

The desktop loader provides a callback that can be used for initializing a dispatchable object. The callback is passed as an extension structure via the pNext field in the create info structure when creating an instance (VkInstanceCreateInfo) or device (VkDeviceCreateInfo). The callback prototype is defined as follows for instance and device callbacks respectively (see vk_layer.h):

VKAPI_ATTR VkResult VKAPI_CALL vkSetInstanceLoaderData(VkInstance instance,
                                                       void *object);
VKAPI_ATTR VkResult VKAPI_CALL vkSetDeviceLoaderData(VkDevice device,
                                                     void *object);

To obtain these callbacks the layer must search through the list of structures pointed to by the "pNext" field in the VkInstanceCreateInfo and VkDeviceCreateInfo parameters to find any callback structures inserted by the loader. The salient details are as follows:

  • For VkInstanceCreateInfo the callback structure pointed to by "pNext" is VkLayerInstanceCreateInfo as defined in include/vulkan/vk_layer.h.
  • A "sType" field in of VK_STRUCTURE_TYPE_LOADER_INSTANCE_CREATE_INFO within VkInstanceCreateInfo parameter indicates a loader structure.
  • Within VkLayerInstanceCreateInfo, the "function" field indicates how the union field "u" should be interpreted.
  • A "function" equal to VK_LOADER_DATA_CALLBACK indicates the "u" field will contain the callback in "pfnSetInstanceLoaderData".
  • For VkDeviceCreateInfo the callback structure pointed to by "pNext" is VkLayerDeviceCreateInfo as defined in include/vulkan/vk_layer.h.
  • A "sType" field in of VK_STRUCTURE_TYPE_LOADER_DEVICE_CREATE_INFO within VkDeviceCreateInfo parameter indicates a loader structure.
  • Within VkLayerDeviceCreateInfo, the "function" field indicates how the union field "u" should be interpreted.
  • A "function" equal to VK_LOADER_DATA_CALLBACK indicates the "u" field will contain the callback in "pfnSetDeviceLoaderData".

Alternatively, if an older loader is being used that doesn't provide these callbacks, the layer may manually initialize the newly created dispatchable object. To fill in the dispatch table pointer in newly created dispatchable object, the layer should copy the dispatch pointer, which is always the first entry in the structure, from an existing parent object of the same level (instance versus device).

For example, if there is a newly created VkCommandBuffer object, then the dispatch pointer from the VkDevice object, which is the parent of the VkCommandBuffer object, should be copied into the newly created object.

Layer Manifest File Format

On Windows and Linux (desktop), the loader uses manifest files to discover layer libraries and layers. The desktop loader doesn't directly query the layer library except during chaining. This is to reduce the likelihood of loading a malicious layer into memory. Instead, details are read from the Manifest file, which are then provided for applications to determine what layers should actually be loaded.

The following section discusses the details of the Layer Manifest JSON file format. The JSON file itself does not have any requirements for naming. The only requirement is that the extension suffix of the file ends with ".json".

Here is an example layer JSON Manifest file with a single layer:

{
   "file_format_version" : "1.0.0",
   "layer": {
       "name": "VK_LAYER_LUNARG_overlay",
       "type": "INSTANCE",
       "library_path": "vkOverlayLayer.dll"
       "api_version" : "1.0.5",
       "implementation_version" : "2",
       "description" : "LunarG HUD layer",
       "functions": {
           "vkNegotiateLoaderLayerInterfaceVersion":
               "OverlayLayer_NegotiateLoaderLayerInterfaceVersion"
       },
       "instance_extensions": [
           {
               "name": "VK_EXT_debug_report",
               "spec_version": "1"
           },
           {
               "name": "VK_VENDOR_ext_x",
               "spec_version": "3"
            }
       ],
       "device_extensions": [
           {
               "name": "VK_EXT_debug_marker",
               "spec_version": "1",
               "entrypoints": ["vkCmdDbgMarkerBegin", "vkCmdDbgMarkerEnd"]
           }
       ],
       "enable_environment": {
           "ENABLE_LAYER_OVERLAY_1": "1"
       }
       "disable_environment": {
           "DISABLE_LAYER_OVERLAY_1": ""
       }
   }
}

Here's a snippet with the changes required to support multiple layers per manifest file:

{
   "file_format_version" : "1.0.1",
   "layers": [
      {
           "name": "VK_LAYER_layer_name1",
           "type": "INSTANCE",
           ...
      },
      {
           "name": "VK_LAYER_layer_name2",
           "type": "INSTANCE",
           ...
      }
   ]
}
JSON Node Description and Notes Introspection Query
"file_format_version" Manifest format major.minor.patch version number. N/A
Supported versions are: 1.0.0, 1.0.1, and 1.1.0.
"layer" The identifier used to group a single layer's information together. vkEnumerateInstanceLayerProperties
"layers" The identifier used to group multiple layers' information together. This requires a minimum Manifest file format version of 1.0.1. vkEnumerateInstanceLayerProperties
"name" The string used to uniquely identify this layer to applications. vkEnumerateInstanceLayerProperties
"type" This field indicates the type of layer. The values can be: GLOBAL, or INSTANCE vkEnumerate*LayerProperties
NOTES: Prior to deprecation, the "type" node was used to indicate which layer chain(s) to activate the layer upon: instance, device, or both. Distinct instance and device layers are deprecated; there are now just layers. Allowable values for type (both before and after deprecation) are "INSTANCE", "GLOBAL" and, "DEVICE." "DEVICE" layers are skipped over by the loader as if they were not found.
"library_path" The "library_path" specifies either a filename, a relative pathname, or a full pathname to a layer shared library file. If "library_path" specifies a relative pathname, it is relative to the path of the JSON manifest file (e.g. for cases when an application provides a layer that is in the same folder hierarchy as the rest of the application files). If "library_path" specifies a filename, the library must live in the system's shared object search path. There are no rules about the name of the layer shared library files other than it should end with the appropriate suffix (".DLL" on Windows, and ".so" on Linux). N/A
"api_version" The major.minor.patch version number of the Vulkan API that the shared library file for the library was built against. For example: 1.0.33. vkEnumerateInstanceLayerProperties
"implementation_version" The version of the layer implemented. If the layer itself has any major changes, this number should change so the loader and/or application can identify it properly. vkEnumerateInstanceLayerProperties
"description" A high-level description of the layer and it's intended use. vkEnumerateInstanceLayerProperties
"functions" OPTIONAL: This section can be used to identify a different function name for the loader to use in place of standard layer interface functions. The "functions" node is required if the layer is using an alternative name for vkNegotiateLoaderLayerInterfaceVersion. vkGet*ProcAddr
"instance_extensions" OPTIONAL: Contains the list of instance extension names supported by this layer. One "instance_extensions" node with an array of one or more elements is required if any instance extensions are supported by a layer, otherwise the node is optional. Each element of the array must have the nodes "name" and "spec_version" which correspond to VkExtensionProperties "extensionName" and "specVersion" respectively. vkEnumerateInstanceExtensionProperties
"device_extensions" OPTIONAL: Contains the list of device extension names supported by this layer. One "device_\extensions" node with an array of one or more elements is required if any device extensions are supported by a layer, otherwise the node is optional. Each element of the array must have the nodes "name" and "spec_version" which correspond to VkExtensionProperties "extensionName" and "specVersion" respectively. Additionally, each element of the array of device extensions must have the node "entrypoints" if the device extension adds Vulkan API functions, otherwise this node is not required. The "entrypoint" node is an array of the names of all entrypoints added by the supported extension. vkEnumerateDeviceExtensionProperties
"enable_environment" Implicit Layers Only - OPTIONAL: Indicates an environment variable used to enable the Implicit Layer (w/ value of 1). This environment variable (which should vary with each "version" of the layer) must be set to the given value or else the implicit layer is not loaded. This is for application environments (e.g. Steam) which want to enable a layer(s) only for applications that they launch, and allows for applications run outside of an application environment to not get that implicit layer(s). N/A
"disable_environment" Implicit Layers Only - **REQUIRED:**Indicates an environment variable used to disable the Implicit Layer (w/ value of 1). In rare cases of an application not working with an implicit layer, the application can set this environment variable (before calling Vulkan functions) in order to "blacklist" the layer. This environment variable (which should vary with each "version" of the layer) must be set (not particularly to any value). If both the "enable_environment" and "disable_environment" variables are set, the implicit layer is disabled. N/A
Layer Manifest File Version History

The current highest supported Layer Manifest file format supported is 1.1.0. Information about each version is detailed in the following sub-sections:

Layer Manifest File Version 1.1.0

Layer Manifest File Version 1.1.0 is tied to changes exposed by the Loader/Layer interface version 2.

  1. Renaming "vkGetInstanceProcAddr" in the "functions" section is deprecated since the loader no longer needs to query the layer about "vkGetInstanceProcAddr" directly. It is now returned during the layer negotiation, so this field will be ignored.
  2. Renaming "vkGetDeviceProcAddr" in the "functions" section is deprecated since the loader no longer needs to query the layer about "vkGetDeviceProcAddr" directly. It too is now returned during the layer negotiation, so this field will be ignored.
  3. Renaming the "vkNegotiateLoaderLayerInterfaceVersion" function is being added to the "functions" section, since this is now the only function the loader needs to query using OS-specific calls.
    • NOTE: This is an optional field and, as the two previous fields, only needed if the layer requires changing the name of the function for some reason.

You do not need to update your layer manifest file if you don't change the names of any of the listed functions.

Layer Manifest File Version 1.0.1

The ability to define multiple layers using the "layers" array was added. This JSON array field can be used when defining a single layer or multiple layers. The "layer" field is still present and valid for a single layer definition.

Layer Manifest File Version 1.0.0

The initial version of the layer manifest file specified the basic format and fields of a layer JSON file. The fields of the 1.0.0 file format include:

  • "file_format_version"
  • "layer"
  • "name"
  • "type"
  • "library_path"
  • "api_version"
  • "implementation_version"
  • "description"
  • "functions"
  • "instance_extensions"
  • "device_extensions"
  • "enable_environment"
  • "disable_environment"

It was also during this time that the value of "DEVICE" was deprecated from the "type" field.

Layer Library Versions

The current Layer Library interface is at version 2. The following sections detail the differences between the various versions.

Layer Library API Version 2

Introduced the concept of loader and layer interface using the new vkNegotiateLoaderLayerInterfaceVersion function. Additionally, it introduced the concept of [Layer Unknown Physical Device Extensions](#layer-unknown-physical-device- extensions) and the associated vk_layerGetPhysicalDeviceProcAddr function. Finally, it changed the manifest file defition to 1.1.0.

Layer Library API Version 1

A layer library supporting interface version 1 had the following behavior:

  1. GetInstanceProcAddr and GetDeviceProcAddr were directly exported
  2. The layer manifest file was able to override the names of the GetInstanceProcAddr and GetDeviceProcAddrfunctions.
Layer Library API Version 0

A layer library supporting interface version 0 must define and export these introspection functions, unrelated to any Vulkan function despite the names, signatures, and other similarities:

  • vkEnumerateInstanceLayerProperties enumerates all layers in a layer library.
  • This function never fails.
  • When a layer library contains only one layer, this function may be an alias to the layer's vkEnumerateInstanceLayerProperties.
  • vkEnumerateInstanceExtensionProperties enumerates instance extensions of layers in a layer library.
  • "pLayerName" is always a valid layer name.
  • This function never fails.
  • When a layer library contains only one layer, this function may be an alias to the layer's vkEnumerateInstanceExtensionProperties.
  • vkEnumerateDeviceLayerProperties enumerates a subset (can be full, proper, or empty subset) of layers in a layer library.
  • "physicalDevice" is always VK_NULL_HANDLE.
  • This function never fails.
  • If a layer is not enumerated by this function, it will not participate in device function interception.
  • vkEnumerateDeviceExtensionProperties enumerates device extensions of layers in a layer library.
  • "physicalDevice" is always VK_NULL_HANDLE.
  • "pLayerName" is always a valid layer name.
  • This function never fails.

It must also define and export these functions once for each layer in the library:

  • <layerName>GetInstanceProcAddr(instance, pName) behaves identically to a layer's vkGetInstanceProcAddr except it is exported.

    When a layer library contains only one layer, this function may alternatively be named vkGetInstanceProcAddr.

  • <layerName>GetDeviceProcAddr behaves identically to a layer's vkGetDeviceProcAddr except it is exported.

    When a layer library contains only one layer, this function may alternatively be named vkGetDeviceProcAddr.

All layers contained within a library must support vk_layer.h. They do not need to implement functions that they do not intercept. They are recommended not to export any functions.



Vulkan Installable Client Driver Interface With the Loader

This section discusses the various requirements for the loader and a Vulkan ICD to properly hand-shake.

ICD Discovery

Vulkan allows multiple drivers each with one or more devices (represented by a Vulkan VkPhysicalDevice object) to be used collectively. The loader is responsible for discovering available Vulkan ICDs on the system. Given a list of available ICDs, the loader can enumerate all the physical devices available for an application and return this information to the application. The process in which the loader discovers the available Installable Client Drivers (ICDs) on a system is platform dependent. Windows, Linux and Android ICD discovery details are listed below.

ICD Manifest File Usage

As with layers, on Windows and Linux systems, JSON formatted manifest files are used to store ICD information. In order to find system-installed drivers, the Vulkan loader will read the JSON files to identify the names and attributes of each driver. One thing you will notice is that ICD Manifest files are much simpler than the corresponding layer Manifest files.

See the Current ICD Manifest File Format section for more details.

ICD Discovery on Windows

In order to find installed ICDs, the Vulkan loader will scan the values in the following Windows registry key:

   HKEY_LOCAL_MACHINE\SOFTWARE\Khronos\Vulkan\Drivers

For 32-bit applications on 64-bit Windows, the loader scan's the 32-bit registry location:

   HKEY_LOCAL_MACHINE\SOFTWARE\WOW6432Node\Khronos\Vulkan\Drivers

The loader will look at the appropriate registry location and check each value listed. If the key is of type DWORD, and it has a value of 0, the loader will open the JSON manifest file specified by the name. Each name must be a full pathname to a text manifest file. The Vulkan loader will attempt to open each manifest file to obtain the information about an ICD's shared library (".dll") file.

For example, let us assume the registry contains the following data:

[HKEY_LOCAL_MACHINE\SOFTWARE\Khronos\Vulkan\Drivers\]

"C:\vendor a\vk_vendora.json"=dword:00000000
"C:\windows\system32\vendorb_vk.json"=dword:00000001
"C:\windows\system32\vendorc_icd.json"=dword:00000000

In this case, the loader will step through each entry, and check the value. If the value is 0, then the loader will attempt to load the file. In this case, the loader will open the first and last listings, but not the middle. This is because the value of 1 for vendorb_vk.json disables the driver.

The Vulkan loader will open each enabled manifest file found to obtain the name or pathname of an ICD shared library (".DLL") file.

See the ICD Manifest File Format section for more details.

ICD Discovery on Linux

In order to find installed ICDs, the Vulkan loader will scan the files in the following Linux directories:

    /usr/local/etc/vulkan/icd.d
    /usr/local/share/vulkan/icd.d
    /etc/vulkan/icd.d
    /usr/share/vulkan/icd.d
    $HOME/.local/share/vulkan/icd.d

The "/usr/local/*" directories can be configured to be other directories at build time.

The typical usage of the directories is indicated in the table below.

Location Details
$HOME/.local/share/vulkan/icd.d $HOME is the current home directory of the application's user id; this path will be ignored for suid programs
"/usr/local/etc/vulkan/icd.d" Directory for locally built ICDs
"/usr/local/share/vulkan/icd.d" Directory for locally built ICDs
"/etc/vulkan/icd.d" Location of ICDs installed from non-Linux-distribution-provided packages
"/usr/share/vulkan/icd.d" Location of ICDs installed from Linux-distribution-provided packages

The Vulkan loader will open each manifest file found to obtain the name or pathname of an ICD shared library (".so") file.

See the ICD Manifest File Format section for more details.

Using Pre-Production ICDs on Windows and Linux

Independent Hardware Vendor (IHV) pre-production ICDs. In some cases, a pre-production ICD may be in an installable package. In other cases, a pre-production ICD may simply be a shared library in the developer's build tree. In this latter case, we want to allow developers to point to such an ICD without modifying the system-installed ICD(s) on their system.

This need is met with the use of the "VK_ICD_FILENAMES" environment variable, which will override the mechanism used for finding system-installed ICDs. In other words, only the ICDs listed in "VK_ICD_FILENAMES" will be used.

The "VK_ICD_FILENAMES" environment variable is a list of ICD manifest files, containing the full path to the ICD JSON Manifest file. This list is colon-separated on Linux, and semi-colon separated on Windows.

Typically, "VK_ICD_FILENAMES" will only contain a full pathname to one info file for a developer-built ICD. A separator (colon or semi-colon) is only used if more than one ICD is listed.

NOTE: On Linux, this environment variable will be ignored for suid programs.

ICD Discovery on Android

The Android loader lives in the system library folder. The location cannot be changed. The loader will load the driver/ICD via hw_get_module with the ID of "vulkan". Due to security policies in Android, none of this can be modified under normal use.

ICD Manifest File Format

The following section discusses the details of the ICD Manifest JSON file format. The JSON file itself does not have any requirements for naming. The only requirement is that the extension suffix of the file ends with ".json".

Here is an example layer JSON Manifest file:

{
   "file_format_version": "1.0.0",
   "ICD": {
      "library_path": "path to ICD library",
      "api_version": "1.0.5"
   }
}
Field Name Field Value
"file_format_version" The JSON format major.minor.patch version number of this file. Currently supported version is 1.0.0.
"ICD" The identifier used to group all ICD information together.
"library_path" The "library_path" specifies either a filename, a relative pathname, or a full pathname to a layer shared library file. If "library_path" specifies a relative pathname, it is relative to the path of the JSON manifest file. If "library_path" specifies a filename, the library must live in the system's shared object search path. There are no rules about the name of the ICD shared library files other than it should end with the appropriate suffix (".DLL" on Windows, and ".so" on Linux).
"api_version" The major.minor.patch version number of the Vulkan API that the shared library files for the ICD was built against. For example: 1.0.33.

NOTE: If the same ICD shared library supports multiple, incompatible versions of text manifest file format versions, it must have separate JSON files for each (all of which may point to the same shared library).

ICD Manifest File Versions

There has only been one version of the ICD manifest files supported. This is version 1.0.0.

ICD Manifest File Version 1.0.0

The initial version of the ICD Manifest file specified the basic format and fields of a layer JSON file. The fields of the 1.0.0 file format include:

  • "file_format_version"
  • "ICD"
  • "library_path"
  • "api_version"

ICD Vulkan Entry-Point Discovery

The Vulkan symbols exported by an ICD must not clash with the loader's exported Vulkan symbols. This could be for several reasons. Because of this, all ICDs must export the following function that is used for discovery of ICD Vulkan entry-points. This entry-point is not a part of the Vulkan API itself, only a private interface between the loader and ICDs for version 1 and higher interfaces.

VKAPI_ATTR PFN_vkVoidFunction VKAPI_CALL vk_icdGetInstanceProcAddr(
                                               VkInstance instance,
                                               const char* pName);

This function has very similar semantics to vkGetInstanceProcAddr. vk_icdGetInstanceProcAddr returns valid function pointers for all the global- level and instance-level Vulkan functions, and also for vkGetDeviceProcAddr. Global-level functions are those which contain no dispatchable object as the first parameter, such as vkCreateInstance and vkEnumerateInstanceExtensionProperties. The ICD must support querying global- level entry-points by calling vk_icdGetInstanceProcAddr with a NULL VkInstance parameter. Instance-level functions are those that have either VkInstance, or VkPhysicalDevice as the first parameter dispatchable object. Both core entry-points and any instance extension entry-points the ICD supports should be available via vk_icdGetInstanceProcAddr. Future Vulkan instance extensions may define and use new instance-level dispatchable objects other than VkInstance and VkPhysicalDevice, in which case extension entry-points using these newly defined dispatchable objects must be queryable via vk_icdGetInstanceProcAddr.

All other Vulkan entry-points must either:

  • NOT be exported directly from the ICD library
  • or NOT use the official Vulkan function names if they are exported

This requirement is for ICD libraries that include other functionality (such as OpenGL) and thus could be loaded by the application prior to when the Vulkan loader library is loaded by the application.

Beware of interposing by dynamic OS library loaders if the official Vulkan names are used. On Linux, if official names are used, the ICD library must be linked with -Bsymbolic.

ICD Unknown Physical Device Extensions

Originally, if the loader was called with vkGetInstanceProcAddr, it would result in the following behavior:

  1. The loader would check if core function:
    • If it was, it would return the function pointer
  2. The loader would check if known extension function:
    • If it was, it would return the function pointer
  3. If the loader knew nothing about it, it would call down using GetInstanceProcAddr
    • If it returned non-NULL, treat it as an unknown logical device command.
    • This meant setting up a generic trampoline function that takes in a VkDevice as the first parameter and adjusting the dispatch table to call the ICD/Layers function after getting the dispatch table from the VkDevice.
  4. If all the above failed, the loader would return NULL to the application.

This caused problems when an ICD attempted to expose new physical device extensions the loader knew nothing about, but an application did. Because the loader knew nothing about it, the loader would get to step 3 in the above process and would treat the function as an unknown logical device command. The problem is, this would create a generic VkDevice trampoline function which, on the first call, would attempt to dereference the VkPhysicalDevice as a VkDevice. This would lead to a crash or corruption.

In order to identify the extension entry-points specific to physical device extensions, the following function can be added to an ICD:

PFN_vkVoidFunction vk_icdGetPhysicalDeviceProcAddr(VkInstance instance,
                                                   const char* pName);

This function behaves similar to vkGetInstanceProcAddr and vkGetDeviceProcAddr except it should only return values for physical device extension entry-points. In this way, it compares "pName" to every physical device function supported in the ICD.

The following rules apply:

  • If it is the name of a physical device function supported by the ICD, the pointer to the ICD's corresponding function should be returned.
  • If it is the name of a valid function which is not a physical device function (i.e. an Instance, Device, or other function implemented by the ICD), then the value of NULL should be returned.
  • If the ICD has no idea what this function is, it should return NULL.

This support is optional and should not be considered a requirement. This is only required if an ICD intends to support some functionality not directly supported by a significant population of loaders in the public. If an ICD does implement this support, it should return the address of its vk_icdGetPhysicalDeviceProcAddr function through the vkGetInstanceProcAddr function.

The new behavior of the loader's vkGetInstanceProcAddr with support for the vk_icdGetPhysicalDeviceProcAddr function is as follows:

  1. Check if core function:
    • If it is, return the function pointer
  2. Check if known instance or device extension function:
    • If it is, return the function pointer
  3. Call the layer/ICD GetPhysicalDeviceProcAddr
    • If it returns non-NULL, return a trampoline to a generic physical device function, and setup a generic terminator which will pass it to the proper ICD.
  4. Call down using GetInstanceProcAddr
    • If it returns non-NULL, treat it as an unknown logical device command. This means setting up a generic trampoline function that takes in a VkDevice as the first parameter and adjusting the dispatch table to call the ICD/Layers function after getting the dispatch table from the VkDevice. Then, return the pointer to corresponding trampoline function.
  5. Return NULL

You can see now, that, if the command gets promoted to core later, it will no longer be setup using vk_icdGetPhysicalDeviceProcAddr. Additionally, if the loader adds direct support for the extension, it will no longer get to step 3, because step 2 will return a valid function pointer. However, the ICD should continue to support the command query via vk_icdGetPhysicalDeviceProcAddr, until at least a Vulkan version bump, because an older loader may still be attempting to use the commands.

ICD Dispatchable Object Creation

As previously covered, the loader requires dispatch tables to be accessible within Vulkan dispatchable objects, such as: VkInstance, VkPhysicalDevice, VkDevice, VkQueue, and VkCommandBuffer. The specific requirements on all dispatchable objects created by ICDs are as follows:

  • All dispatchable objects created by an ICD can be cast to void **
  • The loader will replace the first entry with a pointer to the dispatch table which is owned by the loader. This implies three things for ICD drivers
    1. The ICD must return a pointer for the opaque dispatchable object handle
    2. This pointer points to a regular C structure with the first entry being a pointer.
    • NOTE: For any C++ ICD's that implement VK objects directly as C++ classes.
      • The C++ compiler may put a vtable at offset zero if your class is non- POD due to the use of a virtual function.
      • In this case use a regular C structure (see below).
    1. The loader checks for a magic value (ICD_LOADER_MAGIC) in all the created dispatchable objects, as follows (see include/vulkan/vk_icd.h):
#include "vk_icd.h"

union _VK_LOADER_DATA {
    uintptr loadermagic;
    void *loaderData;
} VK_LOADER_DATA;

vkObj alloc_icd_obj()
{
    vkObj *newObj = alloc_obj();
    ...
    // Initialize pointer to loader's dispatch table with ICD_LOADER_MAGIC

    set_loader_magic_value(newObj);
    ...
    return newObj;
}

Handling KHR Surface Objects in WSI Extensions

Normally, ICDs handle object creation and destruction for various Vulkan objects. The WSI surface extensions for Linux and Windows ("VK_KHR_win32_surface", "VK_KHR_xcb_surface", "VK_KHR_xlib_surface", "VK_KHR_mir_surface", "VK_KHR_wayland_surface", and "VK_KHR_surface") are handled differently. For these extensions, the VkSurfaceKHR object creation and destruction may be handled by either the loader, or an ICD.

If the loader handles the management of the VkSurfaceKHR objects:

  1. The loader will handle the calls to vkCreateXXXSurfaceKHR and vkDestroySurfaceKHR functions without involving the ICDs.
    • Where XXX stands for the Windowing System name:
      • Mir
      • Wayland
      • Xcb
      • Xlib
      • Windows
      • Android
  2. The loader creates a VkIcdSurfaceXXX object for the corresponding vkCreateXXXSurfaceKHR call.
    • The VkIcdSurfaceXXX structures are defined in include/vulkan/vk_icd.h.
  3. ICDs can cast any VkSurfaceKHR object to a pointer to the appropriate VkIcdSurfaceXXX structure.
  4. The first field of all the VkIcdSurfaceXXX structures is a VkIcdSurfaceBase enumerant that indicates whether the surface object is Win32, Xcb, Xlib, Mir, or Wayland.

The ICD may choose to handle VkSurfaceKHR object creation instead. If an ICD desires to handle creating and destroying it must do the following:

  1. Support version 3 or newer of the loader/ICD interface.
  2. Export and handle all functions that take in a VkSurfaceKHR object, including:
    • vkCreateXXXSurfaceKHR
    • vkGetPhysicalDeviceSurfaceSupportKHR
    • vkGetPhysicalDeviceSurfaceCapabilitiesKHR
    • vkGetPhysicalDeviceSurfaceFormatsKHR
    • vkGetPhysicalDeviceSurfacePresentModesKHR
    • vkCreateSwapchainKHR
    • vkDestroySurfaceKHR

Because the VkSurfaceKHR object is an instance-level object, one object can be associated with multiple ICDs. Therefore, when the loader receives the vkCreateXXXSurfaceKHR call, it still creates an internal VkSurfaceIcdXXX object. This object acts as a container for each ICD's version of the VkSurfaceKHR object. If an ICD does not support the creation of its own VkSurfaceKHR object, the loader's container stores a NULL for that ICD. On the otherhand, if the ICD does support VkSurfaceKHR creation, the loader will make the appropriate vkCreateXXXSurfaceKHR call to the ICD, and store the returned pointer in it's container object. The loader then returns the VkSurfaceIcdXXX as a VkSurfaceKHR object back up the call chain. Finally, when the loader receives the vkDestroySurfaceKHR call, it subsequently calls vkDestroySurfaceKHR for each ICD who's internal VkSurfaceKHR object is not NULL. Then the loader destroys the container object before returning.

Loader and ICD Interface Negotiation

Generally, for functions issued by an application, the loader can be viewed as a pass through. That is, the loader generally doesn't modify the functions or their parameters, but simply calls the ICDs entry-point for that function. There are specific additional interface requirements an ICD needs to comply with that are not part of any requirements from the Vulkan specification. These addtional requirements are versioned to allow flexibility in the future.

Windows and Linux ICD Negotiation

Version Negotiation Between Loader and ICDs

All ICDs (supporting interface version 2 or higher) must export the following function that is used for determination of the interface version that will be used. This entry-point is not a part of the Vulkan API itself, only a private interface between the loader and ICDs.

   VKAPI_ATTR VkResult VKAPI_CALL
       vk_icdNegotiateLoaderICDInterfaceVersion(
           uint32_t* pSupportedVersion);

This function allows the loader and ICD to agree on an interface version to use. The "pSupportedVersion" parameter is both an input and output parameter. "pSupportedVersion" is filled in by the loader with the desired latest interface version supported by the loader (typically the latest). The ICD receives this and returns back the version it desires in the same field. Because it is setting up the interface version between the loader and ICD, this should be the first call made by a loader to the ICD (even prior to any calls to vk_icdGetInstanceProcAddr).

If the ICD receiving the call no longer supports the interface version provided by the loader (due to deprecation), then it should report VK_ERROR_INCOMPATIBLE_DRIVER error. Otherwise it sets the value pointed by "pSupportedVersion" to the latest interface version supported by both the ICD and the loader and returns VK_SUCCESS.

The ICD should report VK_SUCCESS in case the loader provided interface version is newer than that supported by the ICD, as it's the loader's responsibility to determine whether it can support the older interface version supported by the ICD. The ICD should also report VK_SUCCESS in the case its interface version is greater than the loader's, but return the loader's version. Thus, upon return of VK_SUCCESS the "pSupportedVersion" will contain the desired interface version to be used by the ICD.

If the loader receives an interface version from the ICD that the loader no longer supports (due to deprecation), or it receives a VK_ERROR_INCOMPATIBLE_DRIVER error instead of VK_SUCCESS, then the loader will treat the ICD as incompatible and will not load it for use. In this case, the application will not see the ICDs vkPhysicalDevice during enumeration.

Interfacing With Legacy ICDs or Loader

If a loader sees that an ICD does not export the vk_icdNegotiateLoaderICDInterfaceVersion function, then the loader assumes the corresponding ICD only supports either interface version 0 or 1.

From the other side of the interface, if an ICD sees a call to vk_icdGetInstanceProcAddr before a call to vk_icdGetLoaderICDInterfaceVersion, then it knows that loader making the calls is a legacy loader supporting version 0 or 1. If the loader calls vk_icdGetInstanceProcAddr first, it supports at least version 1. Otherwise, the loader only supports version 0.

Loader Version 4 Interface Requirements

The major change to version 4 of the loader/ICD interface is the support of [Unknown Physical Device Extensions](#icd-unknown-physical-device- extensions] using the vk_icdGetPhysicalDeviceProcAddr function. This function is purely optional. However, if an ICD supports a Physical Device extension, it must provide a vk_icdGetPhysicalDeviceProcAddr function. Otherwise, the loader will continue to treat any unknown functions as VkDevice functions and cause invalid behavior.

Loader Version 3 Interface Requirements

The primary change that occurred in version 3 of the loader/ICD interface was to allow an ICD to handle creation/destruction of their own KHR_surfaces. Up until this point, the loader created a surface object that was used by all ICDs. However, some ICDs may want to provide their own surface handles. If an ICD chooses to enable this support, it must export support for version 3 of the loader/ICD interface, as well as any Vulkan function that uses a KHR_surface handle, such as:

  • vkCreateXXXSurfaceKHR (where XXX is the platform specific identifier [i.e. vkCreateWin32SurfaceKHR for Windows])
  • vkDestroySurfaceKHR
  • vkCreateSwapchainKHR
  • vkGetPhysicalDeviceSurfaceSupportKHR
  • vkGetPhysicalDeviceSurfaceCapabilitiesKHR
  • vkGetPhysicalDeviceSurfaceFormatsKHR
  • vkGetPhysicalDeviceSurfacePresentModesKHR

An ICD can still choose to not take advantage of this functionality by simply not exposing the above the vkCreateXXXSurfaceKHR and vkDestroySurfaceKHR functions.

Loader Version 2 Interface Requirements

Version 2 interface has requirements in three areas:

  1. ICD Vulkan entry-point discovery,
  2. KHR_surface related requirements in the WSI extensions,
  3. Vulkan dispatchable object creation requirements.
Loader Versions 0 and 1 Interface Requirements

Version 0 and 1 interfaces do not support version negotiation via vk_icdNegotiateLoaderICDInterfaceVersion. ICDs can distinguish version 0 and version 1 interfaces as follows: if the loader calls vk_icdGetInstanceProcAddr first it supports version 1; otherwise the loader only supports version 0.

Version 0 interface does not support vk_icdGetInstanceProcAddr. Version 0 interface requirements for obtaining ICD Vulkan entry-points are as follows:

  • The function vkGetInstanceProcAddr must be exported in the ICD library and returns valid function pointers for all the Vulkan API entry-points.
  • vkCreateInstance must be exported by the ICD library.
  • vkEnumerateInstanceExtensionProperties must be exported by the ICD library.

Additional Notes:

  • The loader will filter out extensions requested in vkCreateInstance and vkCreateDevice before calling into the ICD; Filtering will be of extensions advertised by entities (e.g. layers) different from the ICD in question.
  • The loader will not call the ICD for vkEnumerate\*LayerProperties() as layer properties are obtained from the layer libraries and layer JSON files.
  • If an ICD library author wants to implement a layer, it can do so by having the appropriate layer JSON manifest file refer to the ICD library file.
  • The loader will not call the ICD for vkEnumerate\*ExtensionProperties if "pLayerName" is not equal to NULL.
  • ICDs creating new dispatchable objects via device extensions need to initialize the created dispatchable object. The loader has generic trampoline code for unknown device extensions. This generic trampoline code doesn't initialize the dispatch table within the newly created object. See the Creating New Dispatchable Objects section for more information on how to initialize created dispatchable objects for extensions non known by the loader.

Android ICD Negotiation

The Android loader uses the same protocol for initializing the dispatch table as described above. The only difference is that the Android loader queries layer and extension information directly from the respective libraries and does not use the json manifest files used by the Windows and Linux loaders.

Glossary of Terms

Field Name Field Value
Android Loader The loader designed to work primarily for the Android OS. This is generated from a different code-base than the desktop loader. But, in all important aspects, should be functionally equivalent.
Desktop Loader The loader designed to work on both Windows and Linux. This is generated from a different code-base than the Android loader. But in all important aspects, should be functionally equivalent.
Core Function A function that is already part of the Vulkan core specification and not an extension. For example, vkCreateDevice().
Device Call Chain The call chain of functions followed for device functions. This call chain for a device function is usually as follows: first the application calls into a loader trampoline, then the loader trampoline calls enabled layers, the final layer calls into the ICD specific to the device. See the Dispatch Tables and Call Chains section for more information
Device Function A Device function is any Vulkan function which takes a VkDevice, VkQueue, VkCommandBuffer, or any child of these, as its first parameter. Some Vulkan Device functions are: vkQueueSubmit, vkBeginCommandBuffer, vkCreateEvent. See the Instance Versus Device section for more information.
Discovery The process of the loader searching for ICD and Layer files to setup the internal list of Vulkan objects available. On Windows/Linux, the discovery process typically focuses on searching for Manifest files. While on Android, the process focuses on searching for library files.
Dispatch Table An array of function pointers (including core and possibly extension functions) used to step to the next entity in a call chain. The entity could be the loader, a layer or an ICD. See Dispatch Tables and Call Chains for more information.
Extension A concept of Vulkan used to expand the core Vulkan functionality. Extensions may be IHV-specific, platform-specific, or more broadly available. You should always query if an extension exists, and enable it during vkCreateInstance (if it is an instance extension) or during vkCreateDevice (if it is a device extension).
ICD Acronym for Installable Client Driver. These are drivers that are provided by IHVs to interact with the hardware they provide. See Installable Client Drivers section for more information.
IHV Acronym for an Independent Hardware Vendor. Typically the company that built the underlying hardware technology you are trying to use. A typical examples for a Graphics IHV are: AMD, ARM, Imagination, Intel, Nvidia, Qualcomm, etc.
Instance Call Chain The call chain of functions followed for instance functions. This call chain for an instance function is usually as follows: first the application calls into a loader trampoline, then the loader trampoline calls enabled layers, the final layer calls a loader terminator, and the loader terminator calls all available ICDs. See the Dispatch Tables and Call Chains section for more information
Instance Function An Instance function is any Vulkan function which takes as its first parameter either a VkInstance or a VkPhysicalDevice or nothing at all. Some Vulkan Instance functions are: vkEnumerateInstanceExtensionProperties, vkEnumeratePhysicalDevices, vkCreateInstance, vkDestroyInstance. See the Instance Versus Device section for more information.
Layer Layers are optional components that augment the Vulkan system. They can intercept, evaluate, and modify existing Vulkan functions on their way from the application down to the hardware. See the Layers section for more information.
Loader The middle-ware program which acts as the mediator between Vulkan applications, Vulkan layers and Vulkan drivers. See [The Loader](#the loader) section for more information.
Manifest Files Data files in JSON format used by the desktop loader. These files contain specific information for either a Layer or an ICD.
Terminator Function The last function in the instance call chain above the ICDs and owned by the loader. This function is required in the instance call chain because all instance functionality must be communicated to all ICDs capable of receiving the call. See Dispatch Tables and Call Chains for more information.
Trampoline Function The first function in an instance or device call chain owned by the loader which handles the setup and proper call chain walk using the appropriate dispatch table. On device functions (in the device call chain) this function can actually be skipped. See Dispatch Tables and Call Chains for more information.
WSI Extension Acronym for Windowing System Integration. A Vulkan extension targeting a particular Windowing and designed to interface between the Windowing system and Vulkan. See WSI Extensions for more information.