page.title=Vulkan Validation Layers on Android @jd:body
Most explicit graphics APIs do not perform error-checking, because doing so can result in a performance penalty. Vulkan provides error-checking in a manner that lets you use this feature at development time, but exclude it from the release build of your app, thus avoiding the penalty when it matters most. You do this by enabling validation layers. Validation layers intercept or hook Vulkan entry points for various debug and validation purposes.
Each validation layer can contain definitions for one or more of these entry points, and intercepts the entry points for which it contains definitions. When a validation layer does not define an entry point, the system passes the entry point on to the next layer. Ultimately, an entry point not defined in any layer reaches the driver, the base level, unvalidated.
The Android SDK, NDK, and Vulkan samples include Vulkan validation layers for use during development. You can hook these validation layers into the graphics stack, allowing them to report validation issues. This instrumentation allows you to catch and fix misuses during development.
This page explains how to:
This section explains how to build layers from source. If you have precompiled layers, you can skip this section, and instead read about how to install your layers using Android Studio or from the command line.
NDK Revision 12 and later contains source code for Android validation layers that is known-good, and ready to build. This code resides under the {@code <ndk-root>/sources/third_party/vulkan/src/build-android/generated/gradle-build} directory. This version of the layers should be sufficient for most needs. If so, your next task is to build them. Alternatively, you can pull source code from the Khronos Group repository.
Although we recommend that you use the source code provided with the NDK, you can also pull more recent versions of the source code directly from the GitHub repository belonging to the Khronos Group. To do so, perform the following steps.
$ git clone git@github.com:KhronosGroup/Vulkan-LoaderAndValidationLayers.git
Note: You must have a private SSH key associated with GitHub, or this command fails with a {@code Permission denied (publickey)} message.
$ cd Vulkan-LoaderAndValidationLayers $ git checkout android_layers
$ cd build-android $ ./android-generate
> cd build-android > android-generate.bat
Android Studio builds the validation layers when it builds the rest of the app. This flow makes it easier for you to trace through the layers at runtime. Each layer's source code corresponds to a single Gradle project, which you can specify directly in your Android Studio app. For example, there is a {@code build.gradle} project for threading, and another one for parameter validation.
To integrate layers directory into Android Studio application, perform these steps:
// configure your path to the source code generated on your machine
def layerProjRoot = file('/path/to/ndk-root/.../build-android/generated/gradle-build')
String[] layers = ['threading',
'parameter_validation',
'object_tracker',
'core_validation',
'device_limits',
'image',
'swapchain',
'unique_objects']
for (layer in layers) {
include ":"+ layer
project(":" + layer.toString()).projectDir = new File("${layerProjRoot}/${layer}")
}
android.sources {
main {
jni { ... }
jniLibs {
dependencies {
project ":threading"
project ":parameter_validation"
project ":object_tracker"
project ":core_validation"
project ":device_limits"
project ":image"
project ":swapchain"
project ":unique_objects"
}
}
}
} // android.sources
To build validation layers on Linux or OS X, enter these commands on the command line:
$ cd generated/gradle-build $ # configure SDK and NDK path in local.properties $ gradlew assembleAllDebug
$ ndk-build
To build validation layers on Windows, enter these commands on the command line:
> cd generated\gradle-build > REM configure SDK and NDK path in local.properties > gradlew.bat assembleAllDebug
> ndk-build.cmd
After building the layers, you must provide them to your app. To do so, you must first create a {@code jniLibs} folder in your app's project directory under {@code ./src/main/}, and copy the libs to it. The following example shows how to do this.
$ mkdir ./src/main/jniLibs
The next step depends on whether you are using Gradle or Android makefiles. If you're using Gradle, each built layer resides in its own directory. Consolidate the layers into a single directory, as the following example shows:
$ cp -r .../build-android/generated/gradle-build/threading/build/outputs/native/debug/all/lib/* ./src/main/jniLibs/ $ cp -r .../build-android/generated/gradle-build/parameter_validation/build/outputs/native/debug/all/lib/* ./src/main/jniLibs/ $ cp -r .../build-android/generated/gradle-build/object_tracker/build/outputs/native/debug/all/lib/* ./src/main/jniLibs/ $ cp -r .../build-android/generated/gradle-build/core_validation/build/outputs/native/debug/all/lib/* ./src/main/jniLibs/ $ cp -r .../build-android/generated/gradle-build/device_limits/build/outputs/native/debug/all/lib/* ./src/main/jniLibs/ $ cp -r .../build-android/generated/gradle-build/image/build/outputs/native/debug/all/lib/* ./src/main/jniLibs/ $ cp -r .../build-android/generated/gradle-build/swapchain/build/outputs/native/debug/all/lib/* ./src/main/jniLibs/ $ cp -r .../build-android/generated/gradle-build/unique_objects/build/outputs/native/debug/all/lib/* ./src/main/jniLibs/If you're using Android makefiles, the built layers reside in {@code lib} folders, with one {@code lib} folder under each architecture’s root directory. Consolidate the makefiles under the {@code jniLibs} directory as this example shows:
$ cp -r .../build-android/libs/* ./src/main/jniLibs/
Regardless of whether you build using Gradle or Android makefiles, the build process produces a file structure like the following:
src/main/jniLibs/
arm64-v8a/
libVkLayer_core_validation.so
libVkLayer_device_limits.so
libVkLayer_image.so
libVkLayer_object_tracker.so
libVkLayer_parameter_validation.so
libVkLayer_swapchain.so
libVkLayer_threading.so
libVkLayer_unique_objects.so
armeabi-v7a/
libVkLayer_core_validation.so
...
The following example shows how to verify that your APK contains the validation layers as expected:
$ jar -xvf project.apk ... inflated: lib/arm64-v8a/libVkLayer_threading.so inflated: lib/arm64-v8a/libVkLayer_object_tracker.so inflated: lib/arm64-v8a/libVkLayer_swapchain.so inflated: lib/arm64-v8a/libVkLayer_unique_objects.so inflated: lib/arm64-v8a/libVkLayer_parameter_validation.so inflated: lib/arm64-v8a/libVkLayer_image.so inflated: lib/arm64-v8a/libVkLayer_core_validation.so inflated: lib/arm64-v8a/libVkLayer_device_limits.so ...
The Vulkan API allows an app to enable both instance layers and device layers.
A layer that can intercept Vulkan instance-level entry points is called an instance layer. Instance-level entry points are those with {@code VkInstance} or {@code VkPhysicalDevice} as the first parameter.
You can call {@code vkEnumerateInstanceLayerProperties()} to list the available instance layers and their properties. The system enables instance layers when {@code vkCreateInstace()} executes.
The following code snippet shows how an app can use the Vulkan API to programmatically enable and query an instance layer:
// Get instance layer count using null pointer as last parameter
uint32_t instance_layer_present_count = 0;
vkEnumerateInstanceLayerProperties(&instance_layer_present_count, nullptr);
// Enumerate instance layers with valid pointer in last parameter
VkLayerProperties* layer_props =
(VkLayerProperties*)malloc(instance_layer_present_count * sizeof(VkLayerProperties));
vkEnumerateInstanceLayerProperties(&instance_layer_present_count, layer_props));
// Make sure the desired instance validation layers are available
// NOTE: These are not listed in an arbitrary order. Threading must be
// first, and unique_objects must be last. This is the order they
// will be inserted by the loader.
const char *instance_layers[] = {
"VK_LAYER_GOOGLE_threading",
"VK_LAYER_LUNARG_parameter_validation",
"VK_LAYER_LUNARG_object_tracker",
"VK_LAYER_LUNARG_core_validation",
"VK_LAYER_LUNARG_device_limits",
"VK_LAYER_LUNARG_image",
"VK_LAYER_LUNARG_swapchain",
"VK_LAYER_GOOGLE_unique_objects"
};
uint32_t instance_layer_request_count =
sizeof(instance_layers) / sizeof(instance_layers[0]);
for (uint32_t i = 0; i < instance_layer_request_count; i++) {
bool found = false;
for (uint32_t j = 0; j < instance_layer_present_count; j++) {
if (strcmp(instance_layers[i], layer_props[j].layerName) == 0) {
found = true;
}
}
if (!found) {
error();
}
}
// Pass desired instance layers into vkCreateInstance
VkInstanceCreateInfo instance_info = {};
instance_info.sType = VK_STRUCTURE_TYPE_INSTANCE_CREATE_INFO;
instance_info.enabledLayerCount = instance_layer_request_count;
instance_info.ppEnabledLayerNames = instance_layers;
...
A layer that can intercept device-level entry points is called a device layer. Device-level entry points are those whose first parameter is {@code VkDevice}, {@code VkCommandBuffer}, or {@code VkQueue}. The list of device layers to enable is included in the {@code ppEnabledLayerNames} field of the {@code VkDeviceCreateInfo} struct that the app passes into {@code vkCreateDevice()}.
You can call {@code vkEnumerateDeviceLayerProperties} to list the available layers and their properties. The system enables device layers when it calls {@code vkCreateDevice()}.
The following code snippet shows how an app can use the Vulkan API to programmatically enable a device layer.
// Get device layer count using null as last parameter
uint32_t device_layer_present_count = 0;
vkEnumerateDeviceLayerProperties(&device_layer_present_count, nullptr);
// Enumerate device layers with valid pointer in last parameter
VkLayerProperties* layer_props =
(VkLayerProperties *)malloc(device_layer_present_count * sizeof(VkLayerProperties));
vkEnumerateDeviceLayerProperties(physical_device, device_layer_present_count, layer_props));
// Make sure the desired device validation layers are available
// Ensure threading is first and unique_objects is last!
const char *device_layers[] = {
"VK_LAYER_GOOGLE_threading",
"VK_LAYER_LUNARG_parameter_validation",
"VK_LAYER_LUNARG_object_tracker",
"VK_LAYER_LUNARG_core_validation",
"VK_LAYER_LUNARG_device_limits",
"VK_LAYER_LUNARG_image",
"VK_LAYER_LUNARG_swapchain",
"VK_LAYER_GOOGLE_unique_objects"
};
uint32_t device_layer_request_count =
sizeof(device_layers) / sizeof(device_layers[0]);
for (uint32_t i = 0; i < device_layer_request_count; i++) {
bool found = false;
for (uint32_t j = 0; j < device_layer_present_count; j++) {
if (strcmp(device_layers[i],
layer_props[j].layerName) == 0) {
found = true;
}
}
if (!found) {
error();
}
}
// Pass desired device layers into vkCreateDevice
VkDeviceCreateInfo device_info = {};
device_info.sType = VK_STRUCTURE_TYPE_DEVICE_CREATE_INFO;
device_info.enabledLayerCount = device_layer_request_count;
device_info.ppEnabledLayerNames = device_layers;
...
The Debug Report extension {@code VK_EXT_debug_report} allows your application to control layer behavior when an event occurs.
Before using this extension, you must first make sure that the platform supports it. The following example shows how to check for debug extension support and register a callback if the extension is supported.
// Get the instance extension count
uint32_t inst_ext_count = 0;
vkEnumerateInstanceExtensionProperties(nullptr, &inst_ext_count, nullptr);
// Enumerate the instance extensions
VkExtensionProperties* inst_exts =
(VkExtensionProperties *)malloc(inst_ext_count * sizeof(VkExtensionProperties));
vkEnumerateInstanceExtensionProperties(nullptr, &inst_ext_count, inst_exts);
const char * enabled_inst_exts[16] = {};
uint32_t enabled_inst_ext_count = 0;
// Make sure the debug report extension is available
for (uint32_t i = 0; i < inst_ext_count; i++) {
if (strcmp(inst_exts[i].extensionName,
VK_EXT_DEBUG_REPORT_EXTENSION_NAME) == 0) {
enabled_inst_exts[enabled_inst_ext_count++] =
VK_EXT_DEBUG_REPORT_EXTENSION_NAME;
}
}
if (enabled_inst_ext_count == 0)
return;
// Pass the instance extensions into vkCreateInstance
VkInstanceCreateInfo instance_info = {};
instance_info.sType = VK_STRUCTURE_TYPE_INSTANCE_CREATE_INFO;
instance_info.enabledExtensionCount = enabled_inst_ext_count;
instance_info.ppEnabledExtensionNames = enabled_inst_exts;
PFN_vkCreateDebugReportCallbackEXT vkCreateDebugReportCallbackEXT;
PFN_vkDestroyDebugReportCallbackEXT vkDestroyDebugReportCallbackEXT;
vkCreateDebugReportCallbackEXT = (PFN_vkCreateDebugReportCallbackEXT)
vkGetInstanceProcAddr(instance, "vkCreateDebugReportCallbackEXT");
vkDestroyDebugReportCallbackEXT = (PFN_vkDestroyDebugReportCallbackEXT)
vkGetInstanceProcAddr(instance, "vkDestroyDebugReportCallbackEXT");
assert(vkCreateDebugReportCallbackEXT);
assert(vkDestroyDebugReportCallbackEXT);
// Create the debug callback with desired settings
VkDebugReportCallbackEXT debugReportCallback;
if (vkCreateDebugReportCallbackEXT) {
VkDebugReportCallbackCreateInfoEXT debugReportCallbackCreateInfo;
debugReportCallbackCreateInfo.sType =
VK_STRUCTURE_TYPE_DEBUG_REPORT_CREATE_INFO_EXT;
debugReportCallbackCreateInfo.pNext = NULL;
debugReportCallbackCreateInfo.flags = VK_DEBUG_REPORT_ERROR_BIT_EXT |
VK_DEBUG_REPORT_WARNING_BIT_EXT |
VK_DEBUG_REPORT_PERFORMANCE_WARNING_BIT_EXT;
debugReportCallbackCreateInfo.pfnCallback = DebugReportCallback;
debugReportCallbackCreateInfo.pUserData = NULL;
vkCreateDebugReportCallbackEXT(instance, &debugReportCallbackCreateInfo,
nullptr, &debugReportCallback);
}
// Later, when shutting down Vulkan, call the following
if (vkDestroyDebugReportCallbackEXT) {
vkDestroyDebugReportCallbackEXT(instance, debugReportCallback, nullptr);
}
Once your app has registered and enabled the debug callback, the system routes debugging
messages to a callback that you register. An example of such a callback appears below:
#include <android/log.h>
static VKAPI_ATTR VkBool32 VKAPI_CALL DebugReportCallback(
VkDebugReportFlagsEXT msgFlags,
VkDebugReportObjectTypeEXT objType,
uint64_t srcObject, size_t location,
int32_t msgCode, const char * pLayerPrefix,
const char * pMsg, void * pUserData )
{
if (msgFlags & VK_DEBUG_REPORT_ERROR_BIT_EXT) {
__android_log_print(ANDROID_LOG_ERROR,
"AppName",
"ERROR: [%s] Code %i : %s",
pLayerPrefix, msgCode, pMsg);
} else if (msgFlags & VK_DEBUG_REPORT_WARNING_BIT_EXT) {
__android_log_print(ANDROID_LOG_WARN,
"AppName",
"WARNING: [%s] Code %i : %s",
pLayerPrefix, msgCode, pMsg);
} else if (msgFlags & VK_DEBUG_REPORT_PERFORMANCE_WARNING_BIT_EXT) {
__android_log_print(ANDROID_LOG_WARN,
"AppName",
"PERFORMANCE WARNING: [%s] Code %i : %s",
pLayerPrefix, msgCode, pMsg);
} else if (msgFlags & VK_DEBUG_REPORT_INFORMATION_BIT_EXT) {
__android_log_print(ANDROID_LOG_INFO,
"AppName", "INFO: [%s] Code %i : %s",
pLayerPrefix, msgCode, pMsg);
} else if (msgFlags & VK_DEBUG_REPORT_DEBUG_BIT_EXT) {
__android_log_print(ANDROID_LOG_VERBOSE,
"AppName", "DEBUG: [%s] Code %i : %s",
pLayerPrefix, msgCode, pMsg);
}
// Returning false tells the layer not to stop when the event occurs, so
// they see the same behavior with and without validation layers enabled.
return VK_FALSE;
}