|
| 1 | +## 0.5.5 |
| 2 | + |
| 3 | +- Matlab square left and right division now detects full real tridiagonal coefficient matrices, matching the structure introduced for dense inputs in Matlab R2024a. |
| 4 | +- Tridiagonal systems now use an adjacent-row partial-pivoting LU factorization with compact diagonal storage instead of general dense LU. |
| 5 | +- The tridiagonal runtime supports multiple right-hand sides and dedicated transpose solves, so right division uses the same specialized contract. |
| 6 | +- Exact symmetric positive-definite matrices now use Cholesky factorization and triangular substitution in generated JavaScript and C++17. |
| 7 | +- Symmetric matrices that are not positive definite fall back deterministically to partial-pivoting dense LU without exposing a failed Cholesky result. |
| 8 | +- Square structure selection now follows diagonal, triangular, tridiagonal, symmetric-positive-definite, and dense fallback priority in both output targets. |
| 9 | +- Tridiagonal and Cholesky kernels now participate in the same iterative 1-norm reciprocal-condition estimation used for square-system warnings. |
| 10 | +- Exact singular tridiagonal systems and nearly singular positive-definite systems emit the stable Matlab-style warning and continue with the computed result. |
| 11 | +- The verified matrix structure policy now represents full real-square classification through semantic analysis, MIR, JavaScript LIR, and `cpp` LIR. |
| 12 | +- Generated helper names now describe the real structured-square contract and no longer retain the earlier diagonal/triangular-only helper identity. |
| 13 | +- Added executable left/right tridiagonal, positive-definite, symmetric-indefinite, singular, and nearly singular examples with dual-target output and warning validation. |
| 14 | +- Expanded source-map assertions, verifier corruption checks, architecture checks, and Matlab fuzz seeds for the advanced structure paths. |
| 15 | +- The release performance gate now covers nineteen compile scenarios, including a dedicated advanced structured-square workload. |
| 16 | +- Release validation now covers 216 internal tests, 77 differential cases, 19 performance scenarios, and 90.40% production line coverage. |
| 17 | + |
1 | 18 | ## 0.5.4 |
2 | 19 |
|
3 | 20 | - Matlab square left and right division now detects diagonal, upper-triangular, lower-triangular, and general dense real coefficient matrices at runtime. |
|
0 commit comments