ONNC - 0.9.1 release
Download as PPTX, PDF4 likes679 views
The document discusses ONNX and how it aims to connect deep learning models to different hardware accelerators like CPUs, GPUs, DSPs and DLAs. It explains some assumptions made about the target systems and the role of the compiler. Specifically, it discusses different types of spills that can occur during compilation - compulsory, memory and operator spills. It also talks about different strategies a compiler can take to handle operator spills. Finally, it provides information about contributing to the ONNX project and the release schedule.
ONNC - 0.9.1 release
- 1. connect ONNX to every deep learning accelerator https://onnc.ai https://onnc.ai
- 2. executablility CPU GPU DSP DLA https://onnc.ai
- 3. traditional compiler heterogeneous architecture system (HSA) single architecture PetriNet(1) CFG and DFG target programming model IR type ONNX IR (multiple outputs) three address code (single output) physical feature depends on operand opcode https://onnc.ai
- 4. Assumption of target systems • Accelerators are more effective than processors • Processors are more flexible than accelerators • If the communication cost is less than the computation cost, than the task will reside in accelerator • All tasks start from the top level processor CPU DSP DLA flexible effective https://onnc.ai
- 5. CPU DSP DLA CONV load store load store CONV CONV compulsory spill compulsory spill cost effective flexible Compulsory Spill is easy to implement in the other compiler framework https://onnc.ai
- 6. CPU DSP DLA CONV load store load store X store load Y store load memory spill eliminate them in compiler Memory Spill is what we already have in every compiler framework https://onnc.ai
- 7. Z CPU DSP DLA CONV load store load store X store load store Y load store load W store load Z Z operator spill Operator Spill is totally new and required for every accelerators https://onnc.ai
- 8. What a compiler should do when an operator spill occurs? 1. push the operator to upper device 2. split the operator 3. give up this compilation and retry In many cases, option 3 is the only possible solution https://onnc.ai
- 9. traditional compiler heterogeneous architecture system (HSA) single architecture ITERATIVE sequential target compilation model Lattice D BA C Add D, PassManager will add A and B automatically A B D C topologic sort retry https://onnc.ai
- 10. traditional compiler Limited DLA save 377% in avg. paging systemMemory constraint randnet_ manual/t est2 CaffeNet LeNet yolo9000 AlexNet R-CNN- ilsvrc13 yolov1 FlickrStyl eCaffeNe t VGG_ILSV RC_19_la yer VGG_ILSV RC_16_la yer yolov2- tiny yolov1- tiny Ratio (origin size / new size) 361.25% 263.58% 120.83% 615.86% 312.82% 263.34% 1079.96% 264.32% 554.49% 494.60% 443.97% 408.18% 0.00% 200.00% 400.00% 600.00% 800.00% 1000.00% 1200.00% Ratio (origin size / new size) https://onnc.ai
- 11. Connect to both LLVM and ASIC No porting effort for LLVM compiler Support complex ASIC design https://onnc.ai
- 12. for porters for developers for testers Projects reside in https://repo.onnc.ai The Regression project The Umbrella project https://onnc.ai
- 13. How to contribute https://onnc.ai I have a question I have a wish Ask questions in the mailing list Is the wish specific? Is it a long wish? yes Make an issue no no yes
- 14. https://onnc.ai Current Status 0.9.1 1.0.0 ~8/24 Next release often release; fast Iterate (3~4 weeks a release interval)
- 15. https://onnc.ai Give me 罐罐 and Stars, please https://repo.onnc.ai