Fix bug in gemv c code

[ricardoV94]

Fix bug in handling of row/column matrices in GEMV c_code

Bug was caused by reusing the blas-friendly readjusted strides in the logic to decide whether the call to GEMV should be transposed or not.

Particularly the old +1 in the strides was causing the last error branch to be triggered when the matrix had shape (1, 1). The +1 was supposedly there for the cases of matrix with length 0, but that has it's own branch where the adjusted strides are never used. [That branch is actually useless, and was also removed in this PR]

This bug was introduced in afe934b, before it we were using the unadjusted strides, but that is also suboptimal since strides for length1 dimensions are meaningless and any value is valid by the numpy standard.

The bug showed up in pymc-devs/pymc#7792

This PR also introduces a benchmark test for the special logic in the cases where the gemv is doing a vector-vector dot. This was previously in the C-contiguous branch (which comes after the F-contiguous branch), which is sub-optimal because if a vector is C-contiguous it's also F-contiguous. We should apply in those cases regardless of the values the strides happen to be.

[Copilot]

Pull Request Overview

This PR fixes a bug in the GEMV C code related to the handling of row/column matrices, particularly addressing the erroneous +1 addition to strides for 1×1 matrices. It also introduces additional test coverage, including a benchmark for vector–vector dot product performance.

• Updated stride calculations in the GEMV implementation for row/column matrices
• Added test cases and a performance benchmark for vector dot products
• Updated error messaging and cache versioning in the GEMV code

Reviewed Changes

Copilot reviewed 3 out of 3 changed files in this pull request and generated 1 comment.

File	Description
tests/tensor/test_blas_c.py	Added benchmark test for GEMV vector–vector dot product performance
tests/tensor/test_blas.py	Included additional edge case tests (1×1 matrices)
pytensor/tensor/blas_c.py	Adjusted stride calculation, branch conditions, and cache version

[codecov]

Codecov Report

All modified and coverable lines are covered by tests ✅

Project coverage is 82.10%. Comparing base (5ffe17a) to head (3ff344b).
Report is 6 commits behind head on main.

Additional details and impacted files

@@           Coverage Diff           @@
##             main    #1408   +/-   ##
=======================================
  Coverage   82.10%   82.10%           
=======================================
  Files         208      208           
  Lines       49576    49576           
  Branches     8791     8791           
=======================================
  Hits        40704    40704           
  Misses       6699     6699           
  Partials     2173     2173           

Files with missing lines	Coverage Δ
pytensor/tensor/blas_c.py	92.98% <100.00%> (ø)

🚀 New features to boost your workflow:

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