1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
|
- Block bidiagonal
- Explore having the various implementations of MatrixMult_* internally invoke an interface that can be swapped
* Performance hit for small matrices?
*
-------------------
- Unroll symmetric inverse by minor
- Unroll multiplication for square matrices
- Make QrUpdate more friendly and accessible
For some future Release
- Block SVD
- Block Hessenberg
- Adapt QR-tran into LQ
- Remove QRDecompositionHouseholder?
- Remove chol-block for dense with chol-block64?
* reduce cache misses in invert and see if its faster
- Merge inner triangular solver code
- Require LinearSolverFactory to take in a matrix so it can figure out alg to use?
- Improve cholesky block inverse by taking advantage of symmetry
- Add a function for sorting eigenvalues.
----------------------------------------------------------
- Sparse
- Look into supernodal methods. Apparently they are faster since they expoit dense matrix kernels
- LU
- block
- Cholesky
- unwrap for small matrices. improve accuracy
- improve stability
- Linear Solver
* Iterative
* Add condition(), use Hager's method? pg 132
* Put this new condition into NormOps since it should be much faster
- SVD
- Save up rotators, multiply against each other, then multiply against U and V
- Divide and conquer algorithm
- An implementation that just finds zero singular values
- Bidiagonal decompositions have a lot of inefficient code
- Incremental SVD
- Eigen decomposition
- Divide and conquer algorithm.
- Accurate version of symmetric eigenvalue for 2 by 2
- SVD
- SymmEig
- Fast Matrix Multiply
- hard code cholesky decomposition for small matrices
- hard code symmetric inverse for small matrices
- Matrix Multiplication:
- Try a variant on mult_aux that does the vector mult up to block size then goes down a row.
- Finish vector vector multiply
- Code generator for matrix vector ops
- Add matrix vector multiply
- Auto switch to all of above in CommonOps
|
