17bd59ec80
only call alpha calls if CBLAS version will be called.
2010-02-26 03:23:48 +00:00
9955ea20bd
Added pure cblas implementation of multiplyAdd functions. This does not have autotools support yet, so to enable, goto cpp/Makefile.am, comment out the block concerning GSL/ATLAS, and uncomment the above section enabling just blas.
2010-02-24 18:12:48 +00:00
71089a6354
vector_scale_inplace
2010-02-24 06:12:56 +00:00
547427514f
add save functions for Matrix and Vector to save them to disk
2010-02-24 00:12:10 +00:00
5b6af9e4bd
1.0 versions of BLAS level 2
2010-02-20 23:44:07 +00:00
3247751b5d
Major check-in: there are now two interchangeable implementations of VectorConfig.
...
VectorMap uses a straightforward stl::map of Vectors. It has O(log n)
insert and access, and is fairly fast at both. However, it has high overhead
for arithmetic operations such as +, scale, axpy etc...
VectorBTree uses a functional BTree as a way to access SubVectors
in an ordinary Vector. Inserting is O(n) and much slower, but accessing,
is O(log n) and might be a bit slower than VectorMap. Arithmetic operations
are blindingly fast, however. The cost is it is not as KISS as VectorMap.
Access to vectors is now exclusively via operator[]
Vector access in VectorMap is via a Vector reference
Vector access in VectorBtree is via the SubVector type (see Vector.h)
Feb 16 2010: FD: I made VectorMap the default, because I decided to try
and speed up conjugate gradients by using Sparse FactorGraphs all the way.
2010-02-17 03:29:12 +00:00
5d33eb185c
added SubVector type and some related routines
2010-02-15 23:53:16 +00:00
693e13ef88
added default bool option to svd to sort the singular values and V. the default is true so pass false to avoid sorting
2010-02-14 04:54:39 +00:00
ea3b8083b0
Fixed more tests for NonlinearConstraints, inequality constraints now work.
2010-02-07 01:18:17 +00:00
5554f6fc7e
Faster non-GSL versions of BLAS 2-style calls
2010-01-31 22:56:06 +00:00
efe51c8419
Fixed typo in function name
2010-01-31 21:09:03 +00:00
c266012dcd
Added GSL/ATLAS version of multiplyAdd() and transposeMultiplyAdd()
2010-01-31 18:25:03 +00:00
74dd7128ac
Fixup cholesky decomposition - rename to LLt and RtR to make convention clear
2010-01-31 18:07:29 +00:00
48d2dabe43
Added in-place insertion functions to Matrix, as well as a #define flag to enable or disable GSL code without a reconfigure.
2010-01-31 17:21:07 +00:00
5e4b23df59
Matrix::multiplyAdd and transposeMultiplyAdd are "level 2" BLAS and speed up the numeric part of the code substantially. Alex might be able to speed them up even more by making them use real BLAS code within Matrix.cpp.
2010-01-31 16:04:24 +00:00
4f998e5ecd
Matrix::transposeMultiplyAdd
2010-01-31 02:53:03 +00:00
16c55975c1
Fix inverse_square_root, add cholesky decomposition options
2010-01-22 22:28:03 +00:00
4b778a4e82
fixed a bug in matrix solve function
2010-01-22 06:45:01 +00:00
ac746ccead
Improved performance of updateAb in NoiseModel with GSL/ATLAS. Various other small optimizations were made.
2010-01-21 18:51:59 +00:00
063aa14118
Reimplemented matrix::householder_update using GSL and BLAS, you'll need to edit Makefile.am to enable the new version -- see email
2010-01-21 00:59:33 +00:00
5f588031bc
Merged NoiseQR back into trunk
2010-01-20 18:32:48 +00:00
9c9007920a
Changed the interface on Matrix::column() so that it uses the one in our library (now called column_() ) rather than the boost default. Currently, our implementation just uses the boost default, but this may change due to timing results.
2010-01-20 16:08:14 +00:00
98b143cd22
Switched all loop indices in Matrix/Vector to size_t, with improved results in timeGaussianFactorGraph
2010-01-20 14:42:05 +00:00
6bc0462669
Checked Matrix::collect() with different timing metric, and memcpy has approx 25% speedup over original version. Matrix_range is slower or the same speed as the original version.
2010-01-20 13:45:22 +00:00
55e414cbdd
Tried noalias() for collect(), but no real improvement
2010-01-20 03:25:55 +00:00
03ec3e3e62
Implemented a version of collect() using matrix_range, but it had worse performance than memcpy. New version of vector_scale[column] with new pointer arithmetic implementation for approx 20% improvement.
2010-01-20 02:28:25 +00:00
820ae5682d
Optimized row version of vector_scale with approx 20% improvement
2010-01-20 02:04:25 +00:00
b860f06614
Switched Matrix::collect() over to a memcpy, but the results weren't actually an improvement
2010-01-20 01:42:36 +00:00
f577b27f52
Small optimizations on Matrix::collect(), additional timing capabilities, and now collect() can have dimensions specified to avoid need for lookup.
2010-01-19 21:49:22 +00:00
ea14959835
NoiseModel inhertance structure reversed, all constructors protected, new static "factories"
2010-01-17 17:47:23 +00:00
f0ae2c064e
Merging in NoiseModel, not yet used though
2010-01-16 18:39:39 +00:00
a4d61d2f23
Little performance twiddles that make little difference
2010-01-16 07:22:34 +00:00
91a0fb23df
Weighted pseudo-inverse now takes weights (1/sigma^2). Does not make a lot of performance difference.
2010-01-16 06:25:11 +00:00
fb3e38b161
25% performance increase by improving weighted_eliminate
2010-01-16 04:57:58 +00:00
8d28763339
Vector and Matrix initialization optimization
2010-01-12 16:12:31 +00:00
f1f21a8589
inverted sign of square root to make inv sqrt positive for simple covariance matrices
2010-01-10 13:53:31 +00:00
15bb00683a
transpose backsubstitute on upper-triangular matrix
2009-12-30 13:20:16 +00:00
c43cd425ab
new back-substitution functions that can do unit triangular solves
2009-12-29 13:59:34 +00:00
a7a5e5e816
small performance improvement by using a pointer to address r
2009-12-14 16:56:50 +00:00
be10fc2a90
18% performance boost by using pointer to address destination matrix A
2009-12-11 23:59:37 +00:00
d7c67532bf
Changed format of Matrix::vector_scale() to switch arguments
2009-12-08 21:10:38 +00:00
16da0895a8
square_root_inverse
2009-12-08 20:48:13 +00:00
152fa0069a
fixed expected, actual nomenclature in assert_equal: always use (expected, equal) order
2009-12-08 13:53:33 +00:00
ae9789c9ee
Cleaner printing with tabs after each element
2009-12-02 23:01:49 +00:00
f5fc14c0f4
Fixed bug in revealing rank, even simpler than before.
2009-11-13 06:16:56 +00:00
77a1754b69
Renamed LinearFactor -> GaussianFactor, LinearFactorGraph -> GaussianFactorGraph
2009-11-12 16:16:32 +00:00
f51614813e
Improved weighted eliminate to handle arbitrary linear equality constraints
2009-11-11 14:42:09 +00:00
ddc0173671
Basic handling of constraints now works in factor graphs, assuming there is only one constraint on any given variable.
2009-11-10 04:36:07 +00:00
2c37c94b5d
Replaced the householder transform with the weighted system
...
Removed constrained components from makefile, they will disappear shortly
2009-11-09 21:34:20 +00:00
03985d16f6
Fixed equality checking with NaNs in Matrix
...
removed extraneous print statement in Vector
2009-11-09 16:57:30 +00:00
Frank Dellaert
Alex Cunningham
Kai Ni
Manohar Paluri
justinca
Richard Roberts
Chris Beall
Eohan George