| 1 | 2008-06-05 v1.3.3 |
|---|
| 2 | * fix the design of specializations to modular<double> modular<float> |
|---|
| 3 | * give a proper name to ModularBalanced |
|---|
| 4 | * fix the bugs in the bound computations (Winograd recursion over the |
|---|
| 5 | finite field was too deep) |
|---|
| 6 | * prepare the interface for integrating compressed representation for |
|---|
| 7 | small finite fields |
|---|
| 8 | 2007-09-28 v1.3.2 |
|---|
| 9 | * add routines fgetrs and fgesv (cf LAPACK), for system solving. |
|---|
| 10 | supports rectangular, over/underdetermined systems. |
|---|
| 11 | 2007-08-29 v1.3.1 |
|---|
| 12 | * add the benchmark directory, for automatic benchmarking against GOTO |
|---|
| 13 | and ATLAS BLAS. Adapted from Pascal Giorgi's benchmark system. |
|---|
| 14 | 2007-08-28 v1.3.0 |
|---|
| 15 | * new version of ftrmm ftrsm: ftrsm based on a multicascade algorithm |
|---|
| 16 | reducing the number of modular reductions). Automated generation of each |
|---|
| 17 | of the 48 specializations |
|---|
| 18 | * several bug fixes |
|---|
| 19 | * add regression tests: testeur_fgemm, testeur_lqup and testeur_ftrsm |
|---|
| 20 | 2007-07-05 v1.2.2 |
|---|
| 21 | * add a transposed version of the LQUP decomposition routine |
|---|
| 22 | LUdivine |
|---|
| 23 | * fix many bugs in LUdivine |
|---|
| 24 | * new schedules for Winograd algorithm for matrix multiplication: |
|---|
| 25 | 2 cases depending whether beta = 0 or not, taken form [Huss |
|---|
| 26 | Ledermann & Al. 96] |
|---|
| 27 | * add rowEchelon and ReducedRowEchelon routines + associated tests |
|---|
| 28 | 2007-06-21 v1.2.1 |
|---|
| 29 | * add the use of float BLAS, if the field caradinality is small enough |
|---|
| 30 | * improve genericity: gemm can be use over any field domain (not |
|---|
| 31 | requiring any conversion to a integral representation) |
|---|
| 32 | * add a variant of Winograd's algorithm with less temporaries for |
|---|
| 33 | the operation C = AxB |
|---|
| 34 | * add ColumnEchelon and ReducedColumnEchelon routines, using an |
|---|
| 35 | inplace algorithm, based on the LQUP decompositon of LUdivine |
|---|
| 36 | * add routines ftrtri (replacing invL), ftrtrm. |
|---|
| 37 | * fix bunch of memory leaks in the tests (not yet finished) |
|---|
| 38 | 2007-03-13 v1.1.2 |
|---|
| 39 | * change the genericity system for trsm to detect Field |
|---|
| 40 | implementations over double (compatibility with LinBox) |
|---|
| 41 | 2007-03-11 v1.1.1 |
|---|
| 42 | * complete preconditioning phase for the new Charpoly algorithm |
|---|
| 43 | * new Charpoly algorithm renamed CharpolyArithProg |
|---|
| 44 | * add exception for failure of the LasVegas algrithm |
|---|
| 45 | * default charpoly is now: 2 attempts to CharpolyArithProg, then LUKrylov |
|---|
| 46 | |
|---|
| 47 | 2007-02-27 v1.1.0 |
|---|
| 48 | * change some naming conventions in the directories |
|---|
| 49 | * add a LQUP routine for small dimension (LUdivine_small) and the |
|---|
| 50 | cascading with LUdivine |
|---|
| 51 | * put the bound computations in the same file |
|---|
| 52 | * add dense_generator.C for the generation of random dense |
|---|
| 53 | matrices in tests |
|---|
| 54 | * add the new algorithm for characteristic polynomial (temporarily |
|---|
| 55 | named frobenius) |
|---|
| 56 | |
|---|
| 57 | 2006-08-11 v1.0.1 |
|---|
| 58 | * add the field implementation modular-positive.h, especially for |
|---|
| 59 | p=2 |
|---|
| 60 | * add a the flag 'balanced' to the finite fields modular<double>, |
|---|
| 61 | to switch to the apropriate bound computation (fgemm and trsm) |
|---|
| 62 | * fix a bug in LUDivine LQUP elimination (initialisation of the |
|---|
| 63 | permutation P for N=1 in the terminal case) |
|---|
| 64 | * fix a bug in the determination of the number of recursive levels |
|---|
| 65 | of Winograd Algorithm. |
|---|
| 66 | |
|---|
