Chapel: Distributed arrays compatibility with existing LinearAlgebra procedures

[rahulghangas]

A case by case check where Distributed arrays break the existing codebase

• [x] Vector
• [x] Matrix
• [x] eye
• [x] transpose
• [x] plus
• [x] minus
• [x] times
• [x] elementDiv
• [x] dot(vector, vector)
• [ ] dot(matrix, vector) [fails]
• [ ] dot(matrix, matrix) [fails]
• [x] inner
• [x] outer
• [ ] inv [fails]
• [ ] matPow [fails]
• [x] cross
• [x] diag
  
  
  
  • works but does not use distributed
  
  
• [x] tril
• [x] triu
• [x] isDiag
• [x] isHermitian
• [x] isSymmetric
• [x] isTril
• [x] isTriu
• [x] isSquare
• [x] trace
• [x] lu
• [x] permute
• [x] det
• [x] norm
• [x] solve_tril
• [x] solve_triu
• [x] solve
• [ ] cholesky [fails]
• [ ] eigvalsh [fails]
• [ ] eigh [fails]
• [ ] eigvals [fails]
• [ ] eig [fails]
• [ ] svd [fails]
• [x] jacobi
• [x] kron
• [x] isDenseArr
• [x] isDenseDom
• [x] isLocalArr
• [x] isLocalDom
• [x] isDenseMatrix
• [x] isDefaultSparseDom
• [x] isDefaultSparseArr

[rahulghangas]

@ben-albrecht need label: gsoc: linearalgebra

[LouisJenkinsCS]

Can you add checkboxes for each compatibility and incompatibility you find?

• [x] Checked
• [ ] Unchecked

- [x] Checked
- [ ] Unchecked

[ben-albrecht]

In addition to @LouisJenkinsCS , it'd be helpful to include notes too, e.g.

• [ ] dot(matrix,matrix)
  
  
  
  • Fails with compile-time error message: blah blah
  
  
• [x] dot(matrix, vector)
• [x] dot(vector, vector)
  
  
  
  • Code is not optimized for distributed arrays and could use some performance improvements
  
  

[ben-albrecht]

@rahulghangas - can you split dot out into 3 bullets depending on arg type, e.g. dot(vector, vector) vs dot(matrix, vector) vs dot(matrix,matrix)? Do all 3 cases fail?

[ben-albrecht]

@rahulghangas - did you check any sparse functions in this experiment? I suspect very few of them work with distributed sparse arrays today.

[rahulghangas]

I haven't, will update the list