viable/strict/1788773762: [MPS] Add naive triangular_solve for complex numbers (#195861)
- PyTorch: 621 events in the last 90 days
- PyTorch: 615th Release in the last 90 days
- Previous: earlier the same day · trunk/4a053096707d3eb77d814636c1f1f24fe5df2a33: [dynamo] Preserve is_inference in after-AOT repros (#193565)
What happened
MPSMatrixSolveTriangular is float-only, so add custom Metal kernel for complex64. The impl gains a conjugate flag: triangular_solve/solve_triangular use plain transpose (transpose=True means A^T, verified against CPU), while _cholesky_solve_helper_mps now uses the adjoint (A^H) on its transposed solve, which is required for correct complex cholesky_solve and a no-op for real. Fix bug for non-contiguous out handling…
Summary assembled by rule from the sources below