OpenCL support by NNVM & TVM
summary
Shows NNVM+TVM adding an OpenCL backend to run deep‑learning inference on mobile GPUs and edge ARM boards, with examples compiling a simple multiply‑add and a full ResNet-18 end‑to‑end model into TVM modules and deploying them on Firefly/RK3399 and 96Boards devices. Reports TVM outperforms CPU builds and ARM Compute Library on those platforms, notes OpenCL 1.2 portability issues across vendors, and provides scripts and deployment steps that successfully run inference.
AI-written summary · about
chapters (5)
- 0:00Introduction to OpenCL
- 7:24TVM and NNVM overview
- 13:53TVM module and runtime
- 17:50TVM deployment workflow
- 19:45Performance and status
related talks
26mDeep Learning in OpenCV
25mDeep Dive: Accelerating Neural Networks using RVV and Open Standard Software
20mComputer Vision at the Edge with OpenVINO
10mSupport TVM QNN Flow on RISC-V with SIMD Computation
37mDeep Learning Neural Network Acceleration at the Edge
17mEmpower ML Framework on RISC-V by Backend Graph Compilers
19mOptimizing CNN Model Inference on CPUs
37mDeep Learning Neural Network Acceleration at the Edge