xla
587a6b6d - Update codegen to use xla::Shape (#4111)

Comment changes are shownComment changes are hidden
Commit
2 years ago
Update codegen to use xla::Shape (#4111) * Update GenXlaLazyIR codegen to use xla::Shape * Add xla::Shape to custom shape class * Revert "Update GenXlaLazyIR codegen to use xla::Shape" This reverts commit 513350350c923dfe9a7c0e1657dac90602e9a177. * Update xla codegen to have custom GenLazyNativeFuncDefinition generator * Override build_ir_node() in GenXlaLazyNativeFuncDefinition * Override gen() in GenXlaLazyIr * Update existing MakeNode calls to remove torch::lazy::Shape vector * Remove torch::lazy::Shape vector for _adaptive_avg_pool2d_* ops * Update _adaptive_avg_pool3d_* ops as well * Run linter * Add torch pin * Remove lazy::Shape related logic in XLATensor::Create * Clean up scripts/gen_lazy_tensor.py and add some comments * Update cpp tests to remove torch::lazy:Shape constructor calls * Run linter again * Move bitwise ops to ir_codegen * Use new flag use_lazy_shape in xla lazy codegen * Update torch pin * Clean up test_symint.cpp * Run linter on python files * Delete .torch_pin
Author
Parents
  • scripts
    • File
      gen_lazy_tensor.py
  • test/cpp
    • File
      test_symint.cpp
  • torch_xla/csrc
    • File
      aten_xla_type.cpp
    • ops
      • File
        ops.cpp
    • File
      tensor.cpp
    • File
      tensor.h
    • File
      tensor_methods.cpp
  • File
    xla_native_functions.yaml