星星博客 »  > 

TVM性能评估分析(六)

TVM性能评估分析(六)
在这里插入图片描述

Figure 1. The workflow of development PC, compile, deploy to the device, test, then modify the codes again to see whether it accelerates.
在这里插入图片描述

Figure 2. The Android APP takes shared library as input and runs compiled functions on the mobile phone.
在这里插入图片描述

Figure 3. Build TVM functions and NDArrays on a remote device. The ability to cross-compile to different platforms makes it easy to develop on one platform and test on another.
在这里插入图片描述

Figure 4. The instruction to build for your Android device. Once the APK is built, sign it using apps/android_rpc/dev_tools and install it on the phone.
在这里插入图片描述

Figure 5. The NNVM compiler support of TVM stack, we can now directly compile descriptions from deep learning frameworks and compile them to bare metal code that runs on AMD GPUs.
在这里插入图片描述

Figure 6. With ROCm backend, the generic workflow
在这里插入图片描述

Figure 7. The ONNX library to load the ONNX model into the Protocol buffer object.
在这里插入图片描述

Figure 8. An end to end compilation pipeline from front-end deep learning frameworks to bare metal hardwares.
在这里插入图片描述

Figure 9. Typical workflow of NNVM Compiler
在这里插入图片描述

Figure 10. Separation of Optimization and Deployment
在这里插入图片描述

Figure 11. Time Cost of Inference on K80
在这里插入图片描述

Figure 12. The cost of inference on Raspberry PI

相关文章