[take-cheeze]

MaxPool ignores its padded elements, which is equivalent to padding with -inf. The pass previously folded a Pad with constant_value=0 (or an unspecified value, which defaults to 0) into MaxPool by moving the padding into the pool's pads attribute. This changes the result whenever a pooling window's real values are all negative: Pad(0)+MaxPool yields 0 while the fused MaxPool yields the (negative) window maximum.

Make the required Pad constant value depend on the pool type:

• AveragePool (with count_include_pad=1): 0
• MaxPool: -inf

Also restructure the constant-value check so an unspecified Pad value correctly blocks fusion into MaxPool instead of being treated as a match.

Update the MaxPool tests that encoded the old behavior to assert no fusion for value=0/default, and add companion tests confirming Pad(value=-inf) folds correctly.

Fixes onnxsim/onnxsim#290

Claude-Session: https://claude.ai/code/session_013sB8xKJyd86p47c2vfXMVD

[andife]

Added a signed-off follow-up fix in a79a41c, preserving this PR's original contribution and history.

Preserve observable MaxPool indices: skip Pad fusion when the indices output is consumed or exported, while still allowing an unused second output.

Validation on Windows after rebuilding the native extension against this branch's pinned ONNX:

• python -m pytest -o addopts= onnxoptimizer\test\optimizer_test.py -k "fuse_pad_into_maxpool or fuse_pad_into_avgpool" -q — 35 passed; 3 subtests passed.
• git diff --check — passed.

Full suite and non-Windows builds were not run locally. This follow-up was prepared with LLM assistance and checked with targeted regression tests (see #357).