代码拉取完成，页面将自动刷新
This project contains 5 Vulkan layers:
VK_COMPILE_TIME_LOG environment variable.VK_RUNTIME_LOG environment variable.VK_FRAME_TIME_EXIT_AFTER_FRAME environment variable. The output log file location can be set with the VK_FRAME_TIME_LOG environment variable. Benchmark start detection is controlled by the VK_FRAME_TIME_BENCHMARK_WATCH_FILE (which file to incrementally scan) and VK_FRAME_TIME_BENCHMARK_START_STRING (string that denotes benchmark start) environment variables.VK_PIPELINE_CACHE_SIDELOAD_FILE environment variable. The layer creates an implicit pipeline cache object for each device, initialized with the specified file contents, which then gets merged into application pipeline caches (if any), and makes sure that a valid pipeline cache handle is passed to every pipeline creation. This layer does not produce .csv log files.VK_MEMORY_USAGE_LOG environment variable.The results are saved in the CSV format to the specified files.
We support the following log formats:
Each layer (except the cache sideload layer) has a dedicated CSV file into which writes the logs. The files start with the CSV header on top followed by the data entries written in the subsequent lines. In the Analysis Scripts section, we will explain how a CSV can be used.
CommonFile formatA custom format to have all the logs generated by the layers in a single file. A sample format can be seen in events.log. Other than the logs going to the CSV files, there are other logs in the common file indicating the initialization of a layer, shader creation, etc. All the logs contain timestamps making them suitable for a timeline-based plot. A sample plot is depicted in the Analysis Scripts section.
To specify the output destination, set VK_PERFORMANCE_LAYERS_EVENT_LOG_FILE environment variable. For example:
export VK_PERFORMANCE_LAYERS_EVENT_LOG_FILE=events.log
This format enables us to integrate with tools such as chrome://tracing and Perfetto for a better visualization. Here is an example of this format:
{ "name": "Complete Event", "cat": "pipeline", "ph": "X", "ts": 12345, "pid": 123, "tid": 456, "dur": 123, "args": {...} }
To see more examples, check out trace_event.log.
All the events that go the CommonFile will also go to the TraceEvent file but with a different format. The events are JSON objects grouped in a JSON array.
To specify the output destination, set VK_PERFORMANCE_LAYERS_TRACE_EVENT_LOG_FILE environment variable. For example:
export VK_PERFORMANCE_LAYERS_TRACE_EVENT_LOG_FILE=trace.log
The file can then be imported into chrome://tracing and Perfetto.
Here is a sample output rendered by Perfetto:
For more information about the Chrome Trace Event format see: https://docs.google.com/document/d/1CvAClvFfyA5R-PhYUmn5OOQtYMH4h6I0nSsKchNAySU/preview.
The layers are considered experimental. We welcome contributions and suggestions for improvements; see docs/contributing.md.
The project comes with a set of simple log analysis scripts:
You can find more details in the descriptions included in each script file.
Sample build instructions:
# Checkout the submodules.
git submodule update --init
# Build and install performance layers.
mkdir -p <BUILD_DIR> ; cd <BUILD_DIR>
cmake .. \
-GNinja \
-DCMAKE_C_COMPILER=<COMPILER> \
-DCMAKE_CXX_COMPILER=<COMPILER> \
-DCMAKE_BUILD_TYPE=<CONFIGURATION> \
-DCMAKE_INSTALL_PREFIX=run \
-DVULKAN_HEADERS_INSTALL_DIR=<PATH_TO_VULKAN_HEADERS_INSTALL> \
-DVULKAN_LOADER_GENERATED_DIR=<PATH_TO_VULKAN_LOADER>/loader/generated \
&& ninja \
&& ninja install
CMAKE_BUILD_TYPE options are: Release, Debug, RelWithDebInfo, MinSizeRel.
See docker/build.Dockerfile for detailed Ubuntu build instructions.
DVULKAN_HEADERS_INSTALL_DIR and VULKAN_LOADER_GENERATED_DIR
VULKAN_HEADERS_INSTALL_DIR and VULKAN_LOADER_GENERATED_DIR must be absolute paths.
If you have the Vulkan SDK installed in $VULKAN_SDK, you can simply use
VULKAN_HEADERS_INSTALL_DIR=${VULKAN_SDK}
VULKAN_LOADER_GENERATED_DIR=${VULKAN_SDK}/include/vulkan
VULKAN_HEADERS_INSTALL_DIR must be set to the installation directory of Vulkan-Headers. For example, if your build directory is VULKAN_HEADERS_BUILD and your build commands were:
cmake <PATH_TO_VULKAN_HEADERS> -GNinja -DCMAKE_INSTALL_PREFIX=run
ninja
ninja install
you should set VULKAN_HEADERS_INSTALL_DIR to VULKAN_HEADERS_BUILD/run.
VULKAN_LOADER_GENERATED_DIR should be the directory that contains vk_layer_dispatch_table.h. For example, if you cloned Vulkan-Loader to PATH_TO_VULKAN_LOADER, you should set VULKAN_LOADER_GENERATED_DIR to PATH_TO_VULKAN_LOADER/loader/generated.
For operating systems other than Linux, see: https://vulkan.lunarg.com/doc/view/1.3.211.0/linux/layer_configuration.html or the documentation from your Vulkan SDK vendor.
There are many ways to enable Vulkan layers, and we present one way that works on Linux. For a more comprehensive Vulkan layer configuration tutorial, see https://vulkan.lunarg.com/doc/view/1.3.211.0/linux/layer_configuration.html.
$LD_LIBRARY_PATH: Need to append the directory containing the .so files of performance layers to the $LD_LIBRARY_PATH.
export LD_LIBRARY_PATH=<path-with-layers.so>:$LD_LIBRARY_PATH
$VK_INSTANCE_LAYERS: Specify which of the following layers to enable.
To enable multiple layers, separate them with colons. The following command enables both compile time and runtime layers.
export VK_INSTANCE_LAYERS=VK_LAYER_STADIA_pipeline_compile_time:VK_LAYER_STADIA_pipeline_runtime
VK_LAYER_PATH: Path to the directory containing the json files with the name of the layers.
export VK_LAYER_PATH=<path-with-layer-json>
Each layer writes its results in a specific log file. You can determine the path to the log file for each layer by setting its corresponding environment variable.
export VK_COMPILE_TIME_LOG=<log-file-path>
To check if the layers are enabled, you can run a sample Vulkan application (such as vkcube) and look for the generated layer log files, or check the Vulkan loader logs.
This is not an officially supported Google product. Support and/or new releases may be limited.
此处可能存在不合适展示的内容，页面不予展示。您可通过相关编辑功能自查并修改。
如您确认内容无涉及 不当用语 / 纯广告导流 / 暴力 / 低俗色情 / 侵权 / 盗版 / 虚假 / 无价值内容或违法国家有关法律法规的内容，可点击提交进行申诉，我们将尽快为您处理。
马建仓 AI 助手
