Add HannWindow operator to burn-tensor

[walkinggo]

Pull Request Template

Checklist

• Confirmed that cargo run-checks command has been executed. (Note: Ran targeted checks instead due to local CUDA SDK limitations, see Testing section below for details).
• Made sure the book is up to date with changes in this PR. (Note: N/A, this is an API addition and the Rustdoc is fully documented).

Related Issues/PRs

Addresses part of #4531 (Specifically the HannWindow signal processing operator).

Changes

Added the HannWindow operator to the Tensor API to support signal processing models and align with ONNX specifications.

Detailed implementation:

1. crates/burn-tensor/src/tensor/api/float.rs:

   • Added Tensor<B, 1>::hann_window(size, periodic, options) as an associated function under a new impl<B> Tensor<B, 1> block.
   • Implemented the Hann window formula: $w_n = 0.5 - 0.5 \times \cos(2\pi n / N)$ using existing tensor operations (arange, float, cos, mul_scalar, add_scalar, cast).
   • Supported both periodic ($N = size$) and symmetric ($N = size - 1$) modes.
   • Handled edge cases perfectly: size = 0 returns an empty tensor, and size = 1 returns [1.0] (preventing division-by-zero panics and matching PyTorch/NumPy behavior).
   • Used f64 precision for the angular increment calculation to ensure high accuracy before casting to the target dtype.
   • Added dual-state documentation (cfg_attr) for clean KaTeX math rendering in Rustdoc and readable plain-text fallbacks in IDE hover hints.

2. crates/burn-backend-tests/tests/tensor/float/ops/hann_window.rs & mod.rs:

   • Registered the new test module.
   • Added 6 comprehensive test cases covering periodic mode, symmetric mode, dtype options, empty tensors, and the size=1 boundary for both modes.
   • Used assert_approx_eq with Tolerance for cross-backend floating-point stability.

Testing

Due to my local CUDA SDK being an older version, the cudarc binding in cubecl-cuda failed to find CUtensorMapL2promotion and CUtensorMapFloatOOBfill. Therefore, I had to skip the global cargo run-checks command.

Instead, I manually and successfully verified the code using targeted commands for the active backends:

Check	Command	Result
Formatting	cargo fmt --all -- --check	Passed ✅
Linting	cargo clippy -p burn-tensor -p burn-backend-tests -- -D warnings	Passed ✅
Testing	cargo test -p burn-backend-tests --test tensor -- hann_window	6/6 Passed ✅

I had check every line code to make sure the PR is right

[codecov]

Codecov Report

✅ All modified and coverable lines are covered by tests.
✅ Project coverage is 61.83%. Comparing base (4e962f9) to head (4b22def).
⚠️ Report is 4 commits behind head on main.

❌ Your project check has failed because the head coverage (61.83%) is below the target coverage (80.00%). You can increase the head coverage or adjust the target coverage.

Additional details and impacted files

@@            Coverage Diff             @@
##             main    #4631      +/-   ##
==========================================
+ Coverage   61.71%   61.83%   +0.12%     
==========================================
  Files        1095     1096       +1     
  Lines      142906   142972      +66     
==========================================
+ Hits        88192    88405     +213     
+ Misses      54714    54567     -147     

☔ View full report in Codecov by Sentry.
📢 Have feedback on the report? Share it here.

🚀 New features to boost your workflow:

• ❄️ Test Analytics: Detect flaky tests, report on failures, and find test suite problems.

[walkinggo]

Hi @antimora , I've submitted this PR to address the HannWindow part of your issue #4531
Whenever you have some free time, I would really appreciate it if you could take a look and review this :)

[laggui]

Implementation LGTM!

@antimora I'm not super familiar with these signal processing operations, but maybe we should separate them from the tensor API methods? If they're mostly composed of high-level tensor ops, perhaps we could have a signal module (similar to how we have grid operations grouped).

[antimora]

Currently I am traveling; I'll review carefully when I am back this weekend.

[laggui]

No worries! Safe travels 🙂