Extends VK_ARM_data_graph to enable execution of QNN (Qualcomm Neural Network) models on the Qualcomm AI Engine — either the Hexagon NPU (VK_PHYSICAL_DEVICE_DATA_GRAPH_PROCESSING_ENGINE_TYPE_NEURAL_QCOM) or Adreno GPU — via data graph pipelines. Applications import pre-compiled model binaries as pipeline caches with a new VK_PIPELINE_CACHE_HEADER_VERSION_DATA_GRAPH_QCOM header, or select implementation-provided built-in models. Tensor inputs and outputs interoperate with GPU image processing through tensor/image memory aliasing, enabling end-to-end heterogeneous ML inference workflows within Vulkan.
The task list for the VK_QCOM_data_graph_model release is:
As each component is made public, the task will be checked off. When all tasks have been completed this issue will be closed and the extension will be fully released.
Extends
VK_ARM_data_graphto enable execution of QNN (Qualcomm Neural Network) models on the Qualcomm AI Engine — either the Hexagon NPU (VK_PHYSICAL_DEVICE_DATA_GRAPH_PROCESSING_ENGINE_TYPE_NEURAL_QCOM) or Adreno GPU — via data graph pipelines. Applications import pre-compiled model binaries as pipeline caches with a newVK_PIPELINE_CACHE_HEADER_VERSION_DATA_GRAPH_QCOMheader, or select implementation-provided built-in models. Tensor inputs and outputs interoperate with GPU image processing through tensor/image memory aliasing, enabling end-to-end heterogeneous ML inference workflows within Vulkan.The task list for the VK_QCOM_data_graph_model release is:
As each component is made public, the task will be checked off. When all tasks have been completed this issue will be closed and the extension will be fully released.