What type of PR is this?

/kind test

What does this PR do / why do we need it?

This PR adds tests for ShapeEnv APIs that do not have direct upstream test coverage in PyTorch v2.10.0.

Covered APIs:

• torch.fx.experimental.symbolic_shapes.ShapeEnv.get_pruned_guards
• torch.fx.experimental.symbolic_shapes.ShapeEnv.is_unbacked_symint

These APIs are Python-level symbolic shape utilities. They are used for ShapeEnv guard pruning and unbacked SymInt identification, and do not involve NPU operator kernels, device-side numerical computation, stream/event management, memory management, or distributed communication.

Which issue(s) this PR fixes?

Related to Torch-NPU API completion work for ShapeEnv APIs.

Modification summary

Modified file:

test/fx/test_symbolic_shapes.py

Changes:

• Added import sympy.
• Added test_shape_env_get_pruned_guards to verify ShapeEnv.get_pruned_guards returns a list for unbacked SymInt input.
• Added test_shape_env_is_unbacked_symint to verify ShapeEnv.is_unbacked_symint returns True for symbols created by create_unbacked_symint, and False for a regular SymPy symbol.

Upstream test coverage analysis

PyTorch v2.10.0 official tests were checked for the target APIs.

Result:

Therefore, this PR adds Torch-NPU self-written tests for these two APIs.

Relationship with v2.9.0 work

This PR follows the same test strategy as the v2.9.0 ShapeEnv self-written test PR:

https://gitcode.com/Ascend/pytorch/pull/36356

Test result

Attempted to run:

TORCH_DEVICE_BACKEND_AUTOLOAD=0 python3.11 test/fx/test_symbolic_shapes.py

Result:

Traceback (most recent call last):
  File "/workspace/pytorch/test/fx/test_symbolic_shapes.py", line 5, in <module>
    import torch_npu
  File "/workspace/pytorch/torch_npu/__init__.py", line 15, in <module>
    import torch_npu.utils.patch_getenv
  File "/workspace/pytorch/torch_npu/utils/__init__.py", line 12, in <module>
    from torch_npu.npu.utils import get_cann_version
  File "/workspace/pytorch/torch_npu/npu/__init__.py", line 158, in <module>
    from .utils import (obfuscation_initialize, obfuscation_calculate, obfuscation_finalize,
  File "/workspace/pytorch/torch_npu/npu/utils.py", line 11, in <module>
    import torch_npu._C
ModuleNotFoundError: No module named 'torch_npu._C'

Explanation:

The failure occurs during import torch_npu, before entering the added test cases. The current Web IDE environment is a Torch-NPU source checkout without the built torch_npu._C extension loaded. This is an environment/build artifact issue, not a failure of the added ShapeEnv test logic.

Special notes for reviewers

• This PR only adds test coverage for APIs without direct upstream coverage.
• It does not modify API implementation.
• It does not involve NPU kernel behavior or numerical computation.
• ShapeEnv.produce_guards and ShapeEnv.get_nontrivial_guards are handled separately through an upstream test adaptation patch.
• ShapeEnv.ignore_fresh_unbacked_symbols already has upstream coverage and is device-independent; for v2.9.0 this was explained in issue #2046.