Google Git
Sign in
android / platform / external / pytorch / cd449e260c / . / torch / csrc / jit / passes / onnx
tree: 948bafd2132cd576b2afb3f41a89132366463630 [path history] [tgz]
  1. pattern_conversion/
  2. cast_all_constant_to_floating.cpp
  3. cast_all_constant_to_floating.h
  4. constant_fold.cpp
  5. constant_fold.h
  6. constant_map.cpp
  7. constant_map.h
  8. deduplicate_initializers.cpp
  9. deduplicate_initializers.h
  10. eliminate_unused_items.cpp
  11. eliminate_unused_items.h
  12. eval_peephole.cpp
  13. eval_peephole.h
  14. fixup_onnx_controlflow.cpp
  15. fixup_onnx_controlflow.h
  16. function_extraction.cpp
  17. function_extraction.h
  18. function_substitution.cpp
  19. function_substitution.h
  20. helper.cpp
  21. helper.h
  22. list_model_parameters.cpp
  23. list_model_parameters.h
  24. naming.cpp
  25. naming.h
  26. onnx_log.cpp
  27. onnx_log.h
  28. peephole.cpp
  29. peephole.h
  30. prepare_division_for_onnx.cpp
  31. prepare_division_for_onnx.h
  32. preprocess_for_onnx.cpp
  33. preprocess_for_onnx.h
  34. README.md
  35. remove_inplace_ops_for_onnx.cpp
  36. remove_inplace_ops_for_onnx.h
  37. scalar_type_analysis.cpp
  38. scalar_type_analysis.h
  39. shape_type_inference.cpp
  40. shape_type_inference.h
  41. unpack_quantized_weights.cpp
  42. unpack_quantized_weights.h
torch/csrc/jit/passes/onnx/README.md

The optimization passes in this directory work exclusively on ONNX-style IRs, e.g., IRs that have had ToONNX applied to them. ONNX defines operators differently from ATen, so there are different opportunities for peephole optimization.