Add size divisibility specialization

[voltjia]

Summary

• Pass int64:16 divisibility hints to Triton for size parameters whose dim is divisible by 16, enabling vectorized loads alongside the existing stride-1 contiguity hint.
• Extend the C++ dispatcher to pick the most-specific variant by also checking shape[dim] % 16 == 0, falling back to less-specialized variants otherwise.
• Rename the existing stride-spec vocabulary to contiguity so the new divisibility axis sits alongside it without ambiguity (variant suffixes go from s_<dims> to divisibility_<...>_contiguity_<...>).

Testing

pytest output:

============================= test session starts ==============================
platform linux -- Python 3.10.16, pytest-9.0.2, pluggy-1.6.0
rootdir: /home/huangjiacheng/ninetoothed
configfile: pyproject.toml
plugins: anyio-4.12.1, xdist-3.8.0, cov-7.0.0, typeguard-4.4.4
collected 213 items

tests/test_add.py .                                                      [  0%]
tests/test_addmm.py ..                                                   [  1%]
tests/test_aot.py .........                                              [  5%]
tests/test_aot_auto_tuning.py ....                                       [  7%]
tests/test_attention.py ........                                         [ 11%]
tests/test_auto_tuner.py ....                                            [ 13%]
tests/test_clone.py ....                                                 [ 15%]
tests/test_conv2d.py ....                                                [ 16%]
tests/test_data_ptr.py .                                                 [ 17%]
tests/test_debugging.py .                                                [ 17%]
tests/test_dropout.py .                                                  [ 18%]
tests/test_eval.py ........                                              [ 22%]
tests/test_expand.py .                                                   [ 22%]
tests/test_generation.py ............................................... [ 44%]
.............................                                            [ 58%]
tests/test_getitem.py ..........                                         [ 62%]
tests/test_ipynb.py .                                                    [ 63%]
tests/test_jagged.py ................                                    [ 70%]
tests/test_matmul.py ..                                                  [ 71%]
tests/test_max_pool2d.py ..                                              [ 72%]
tests/test_naming.py .......                                             [ 76%]
tests/test_pad.py ................................................       [ 98%]
tests/test_pow.py .                                                      [ 99%]
tests/test_softmax.py .                                                  [ 99%]
tests/test_unsqueeze.py .                                                [100%]

======================= 213 passed in 3161.00s (0:52:41) =======================

[chatgpt-codex-connector]

💡 Codex Review

Here are some automated review suggestions for this pull request.

Reviewed commit: 3ebd64b6dd

ℹ️ About Codex in GitHub

Your team has set up Codex to review pull requests in this repo. Reviews are triggered when you

• Open a pull request for review
• Mark a draft as ready
• Comment "@codex review".

If Codex has suggestions, it will comment; otherwise it will react with 👍.

Codex can also answer questions or update the PR. Try commenting "@codex address that feedback".