Create a Vulkan Instance

Code file for this section is 01-init_instance.cpp

The first step in the samples progression is creating a Vulkan instance. Find the 01-init_instance.cpp program in the API-Samples folder of the LunarG Vulkan Sample repository and get ready to look at it as you read this section about Vulkan instances.

Vulkan Instances

The Vulkan API uses the vkInstance object to store all per-application state. The application must create a Vulkan instance before performing any other Vulkan operations.

The basic Vulkan architecture looks like:

Basic App Loader

The above diagram shows that a Vulkan application is linked to a Vulkan library which is commonly referred to as the loader. Creating an instance initializes the loader. The loader also loads and initializes the low-level graphics driver, usually provided by the vendor of the GPU hardware.

Note that there are layers depicted in this diagram, which are also loaded by the loader. Layers are commonly used for validation, which is the error checking that would normally be performed by the driver. In Vulkan, the drivers are much more lightweight than in other APIs such as OpenGL partly because they delegate this validation function to the validation layers. Layers are optional and can be loaded selectively each time the application creates an instance.

Vulkan layers are outside the scope of this tutorial and of the samples progression. So this tutorial does not cover the layers further. Further information about layers can be found at the Vulkan SDK download site.


Take a look at the 01-init_instance.cpp source code and find the vkCreateInstance function call, which has the following prototype:

VkResult vkCreateInstance(
    const VkInstanceCreateInfo*                 pCreateInfo,
    const VkAllocationCallbacks*                pAllocator,
    VkInstance*                                 pInstance);

Taking this apart, bit by bit:

VkResult - This is the return status of the function. You may wish to open up the vulkan.h header file and inspect some of these definitions as we encounter them. You can find the file in the include directory within the VulkanSamples repository.

VkInstanceCreateInfo - This structure contains any additional information that is needed to create the instance. Since this is an important item, you will take a closer look at it later.

VkAllocationCallbacks - Functions that allocate host memory are equipped with an argument that allows the application to perform its own host memory management. Otherwise, the Vulkan implementation uses the default system memory management facilities. An application might want to manage its own host memory in order to log memory allocations, for example.

The samples do not use this feature, so you will always see NULL passed for this argument in this and other functions throughout the samples.

VkInstance - This is simply a handle returned by this function if the instance creation was successful. This is an opaque handle, so do not try to de-reference it. Many Vulkan functions that create objects return handles for the objects they create in this manner.

A Closer Look at the VkInstanceCreateInfo Structure

Vulkan functions that create objects usually have a VkObjectCreateInfo argument. Find the code in the sample that initializes this structure:

typedef struct VkInstanceCreateInfo {
    VkStructureType             sType;
    const void*                 pNext;
    VkInstanceCreateFlags       flags;
    const VkApplicationInfo*    pApplicationInfo;
    uint32_t                    enabledLayerCount;
    const char* const*          ppEnabledLayerNames;
    uint32_t                    enabledExtensionCount;
    const char* const*          ppEnabledExtensionNames;
} VkInstanceCreateInfo;

The first two members are commonly found in many Vulkan CreateInfo structures.

sType - The sType field indicates the type of the structure. In this case, you set it to VK_STRUCTURE_TYPE_INSTANCE_CREATE_INFO, since it is an VkInstanceCreateInfo structure. This may seem redundant since only a structure of this type can be passed as the first argument of vkCreateInstance(). But it has some value for the following reasons:

Since this member is always the first member in the structure, the consumer can easily determine the type of the structure and decide how to handle it.

pNext - You set pNext to NULL more often than not. This void pointer is sometimes used to pass extension-specific information in a typed structure where the sType member is set to an extension-defined value. As discussed above, extensions can analyze any structures that are passed in along this chain of pNext pointers to find structures that they recognize.

flags - There are currently no flags defined, so set this to zero.

pApplicationInfo - This is a pointer to another structure that you also need to fill out. You will come back to this one in a moment.

enabledLayerCount and ppEnabledLayerNames - The samples in this tutorial do not use layers, so these members are cleared.

enabledExtensionCount and ppEnabledExtensionNames - At this point, the samples in this tutorial do not use extensions. Later on, another sample will show the usage of extensions.

The VkApplicationInfo Structure

This structure provides the Vulkan implementation with some basic information about the application:

typedef struct VkApplicationInfo {
    VkStructureType    sType;
    const void*        pNext;
    const char*        pApplicationName;
    uint32_t           applicationVersion;
    const char*        pEngineName;
    uint32_t           engineVersion;
    uint32_t           apiVersion;
} VkApplicationInfo;

sType and pNext - These have the same meanings as those in the vkInstanceCreateInfo structure.

pApplicationName, applicationVersion, pEngineName, engineVersion - These are free-form fields that the application may provide if desired. Some implementations of tools, loaders, layers, or drivers may use these fields to provide information while debugging or collecting information for reports, etc. It may even be possible for drivers to change their behaviors depending on the application that is running.

apiVersion - This field communicates the major, minor, and patch levels of the Vulkan API header used to compile the application. If you are using Vulkan 1.0, major should be 1 and minor should be 0. Using the VK_API_VERSION_1_0 macro from vulkan.h accomplishes this, with a patch level of 0. Differences in the patch level should not affect the full compatibility between versions that differ only in the patch level. Generally, you should set this field to VK_API_VERSION_1_0 unless you have a good reason to do otherwise.

Back to the Code

Once the structures are populated, the sample app creates the instance:

VkInstance inst;
VkResult res;

res = vkCreateInstance(&inst_info, NULL, &inst);
    std::cout << "cannot find a compatible Vulkan ICD\n";
} else if (res) {
    std::cout << "unknown error\n";

vkDestroyInstance(inst, NULL);

In the code above, the application makes a quick check for the mostly likely error and reports it, or some other error, according to the result. Note that a success (VK_SUCCESS) has a value of zero, so many applications take the shortcut of interpreting a non-zero result as an error, which is the case here.

Finally, the application destroys the instance before it exits.

Now that you have created a Vulkan instance, it is now time to discover what graphics devices are available to your Vulkan instance.

Previous: Introduction Back to: Index Next: Enumerate