This series implements asynchronous model execution for the LibTorch backend. It introduces a dedicated worker thread and a task queue to prevent the main filter thread from blocking during inference, aligning the Torch backend functionality with the OpenVINO backend.
v2 changes: - Split the implementation into 3 patches for better reviewability. - Added proper thread joining and NULL pointer safety in dnn_free_model_th. - Cleaned up comments to follow FFmpeg Doxygen standards. Raja Rathour (3): avfilter/dnn_backend_torch: add async infrastructure avfilter/dnn_backend_torch: implement async execution logic avfilter/dnn_backend_torch: handle lifecycle of worker thread libavfilter/dnn/dnn_backend_torch.cpp | 128 +++++++++++++++++++++----- 1 file changed, 105 insertions(+), 23 deletions(-) -- 2.48.1 _______________________________________________ ffmpeg-devel mailing list -- [email protected] To unsubscribe send an email to [email protected]
