docs: Cleanup GPU-AV docs #8831
Conversation
|
CI Vulkan-ValidationLayers build queued with queue ID 298676. |
|
CI Vulkan-ValidationLayers build # 17957 running. |
|
CI Vulkan-ValidationLayers build # 17957 failed. |
|
CI Vulkan-ValidationLayers build queued with queue ID 299152. |
|
CI Vulkan-ValidationLayers build # 17966 running. |
|
CI Vulkan-ValidationLayers build # 17966 passed. |
| Therefore, the layer keeps a "counter" in per-device state that is incremented each time a shader is instrumented | ||
| to generate unique IDs. | ||
| This unique ID is given to the SPIR-V optimizer and is stored in the shader module state tracker after the shader module is created, which creates the necessary association between the ID and the shader module. | ||
| GPU-AV has many things it validates, all which have different implementation designs. |
There was a problem hiding this comment.
| GPU-AV has many things it validates, all which have different implementation designs. | |
| GPU-AV has many things it validates, all of which have different implementation designs. |
| - A Descriptor Slot | ||
| - GPU-AV requires one descriptor set, which means if an application is using all sets in `VkPhysicalDeviceLimits::maxBoundDescriptorSets`, GPU-AV will not work. | ||
| - There is a `VK_LAYER_GPUAV_RESERVE_BINDING_SLOT` that will reduce `maxBoundDescriptorSets` by one if the app uses that value to determine its logic. | ||
| - `fragmentStoresAndAtomics` and `vertexPipelineStoresAndAtomics` so we can write out information into those stages. |
There was a problem hiding this comment.
| - `fragmentStoresAndAtomics` and `vertexPipelineStoresAndAtomics` so we can write out information into those stages. | |
| - `fragmentStoresAndAtomics` and `vertexPipelineStoresAndAtomics` so we can write out information from those stages. |
| @@ -8,930 +8,59 @@ | |||
|
|
|||
| # GPU Assisted Validation | |||
|
|
|||
| GPU Assisted Validation is implemented in the SPIR-V Tools optimizer and the `VK_LAYER_KHRONOS_validation` layer. | |||
| This document covers the design of the layer portion of the implementation. | |||
| GPU Assisted Validation (GPU-AV) is the part of the Validation Layers that uses the GPU and it's output to validate. | |||
There was a problem hiding this comment.
GPU Assisted Validation
While most validation can be done on the CPU, some things like the content of a buffer or how a shader invocation accesses a descriptor array cannot be known until a command buffer is executed. The validation layers have a dedicated tool to perform validation in those cases: GPU Assisted Validation, or GPU-AV.
GPU-AV directly inspects GPU resources and instrument shaders to validate their run time invocations, and reports back any error. Due to dedicated state tracking, shader instrumentation, necessity to read back data from the GPU, etc... there is a performance overhead, so GPU-AV is off by default.
The current documentation is completely out-dated since the reworking we started last year.
Going through it, I removed a lot as it was to much detail in a doc file and easily was not getting updated
My goal for the future is
docsfolder for how each part of GPU-AV works and point to it from here