pytorch
4a8382b5 - Update caching of tensor arguments for nvFuser's fusion creation (#87860)

Comment changes are shownComment changes are hidden
Commit
2 years ago
Update caching of tensor arguments for nvFuser's fusion creation (#87860) Previously nvFuser's fusion definition was cached based on concrete shape and strides of tensor inputs for simplicity and correctness. This PR changes Python's cache to check the number of dimensions, size-1 dimensions, and contiguity information based on given strides and shapes. Pull Request resolved: https://github.com/pytorch/pytorch/pull/87860 Approved by: https://github.com/kevinstephano, https://github.com/jjsjann123, https://github.com/ngimel
Author
Committer
Parents
  • test
    • File
      test_prims.py
  • torch
    • _prims
      • File
        nvfuser_executor.py
    • csrc/jit/codegen/cuda/python_frontend
      • File
        python_bindings.cpp
Loading