mirror/eigen
2
0
Fork 0
mirror of https://gitlab.com/libeigen/eigen.git synced 2024-12-27 07:29:52 +08:00
Code Issues Packages Projects Releases Wiki Activity
1,561 Commits 11 Branches 89 Tags 168 MiB
d040c9fc9e
Commit Graph

1561 Commits

This Branch
This Branch
All Branches
Author SHA1 Message Date
Gael Guennebaud
13e95f7f68 optimize "apply Jacobi" for small sizes, and move it to Jacobi.h 2009-08-14 00:17:14 +02:00
Benoit Jacob
f2536416da * remove EIGEN_DONT_INLINE that harm performance for small sizes 
* normalize left Jacobi rotations to avoid having to swap rows
* set precision to 2*machine_epsilon instead of machine_epsilon, we lose 1 bit of precision
  but gain between 10% and 100% speed, plus reduce the risk that some day we hit a bad matrix
  where it's impossible to approach machine precision
 2009-08-13 14:56:39 -04:00
Benoit Jacob
76a3089a43 oops, don't set the precision to -1 !! 2009-08-13 09:56:53 -04:00
Benoit Jacob
13b191d94b apply Koldo's workaround for MSVC bug 2009-08-13 09:53:47 -04:00
Gael Guennebaud
1b257a7620 add an optimized "apply in place a rotation in the plane", 
and make Jacobi and SelfAdjointEigenSolver use it
=> ~ x1.75 speedup for JacobiSVD and x2 for SelfAdjointEigenSolver
 2009-08-13 11:42:02 +02:00
Benoit Jacob
1d80f561ad apply change discussed on the list : 
* new default precision "-1" means use the current stream precision
* otherwise, save and restore the stream precision
 2009-08-13 22:50:55 -04:00
Benoit Jacob
99802094e4 do without an empirical homemade formula that i wasn't comfortable about... 
turns out it's not needed anymore and removing it seems to only increase the precision
 2009-08-12 18:30:37 -04:00
Benoit Jacob
6bb6810c5e merge 2009-08-12 18:25:13 -04:00
Benoit Jacob
2b618a2c16 make jacobi SVD more robust after experimenting with very nasty matrices... 
it turns out to be better to repeat the jacobi steps on a given (p,q) pair until it
is diagonal to machine precision, before going to the next (p,q) pair. it's also
an optimization as experiments show that in a majority of cases this allows to find out
that the (p,q) pair is already diagonal to machine precision.
 2009-08-12 18:23:39 -04:00
Jitse Niesen
f71f878bab Add support for matrix exponential of floats and complex numbers. 2009-08-12 15:44:22 +01:00
Benoit Jacob
309d540d4a add parentheses; hopefully this solves Koldos MSVC compilation issue... 2009-08-12 10:14:15 -04:00
Benoit Jacob
22d65d47d0 finally, the good approach was two-sided Jacobi. Indeed, it allows 
to guarantee the precision of the output, which is very valuable.
Here, we guarantee that the diagonal matrix returned by the SVD is
actually diagonal, to machine precision.

Performance isn't bad at all at 50% of the current householder SVD
performance for a 200x200 matrix (no vectorization) and we have
lots of room for improvement.
 2009-08-12 02:35:07 -04:00
Benoit Jacob
ce033ebdfe add EIGEN_DEBUG_VAR 2009-08-11 16:12:34 -04:00
Gael Guennebaud
afbd73b5cd overload operartor* with a ProductBase such that "scalar * (mat * mat)" is optimized 
as one could naturally expect
 2009-08-11 15:15:06 +02:00
Gael Guennebaud
a4f6642518 fix issue #36 (missing return *this in Rotation2D 2009-08-11 15:11:47 +02:00
Gael Guennebaud
ea884e6f48 remove #include Bidiagonalization, and add missing ";" 2009-08-11 15:08:03 +02:00
Gael Guennebaud
35b4077a5d merge 2009-08-09 23:11:25 +02:00
Gael Guennebaud
ef55e7f4ce make custom asm directive volatile 2009-08-09 23:09:46 +02:00
Benoit Jacob
216ee335ac LinearVectorization: If the destination isn't aligned, 
we have to do runtime checks and we don't unroll, so it's only good for large enough sizes
 2009-08-09 22:19:12 +02:00
Benoit Jacob
1f1705868b now you can #define EIGEN_DEBUG_ASSIGN, and all the values in ei_assign_traits are printed 2009-08-09 21:35:13 +02:00
Benoit Jacob
527557672a disable the assembly for fast unaligned stores. indeed, there is a strange bug that is triggered by this code: 
#include<Eigen/Core>

int main()
{
  Eigen::Matrix4f m;
  m <<1,2,3,4,5,6,7,8,9,10,11,12,13,14,15,16;
  m.col(0).swap(m.col(1));
  std::cout << m << std::endl;
}

when the fast unaligned stores are used, the column is copied instead of being swapped.
 2009-08-09 20:49:55 +02:00
Benoit Jacob
8e08680119 don't depend on uninitialized value 2009-08-09 17:18:42 +02:00
Benoit Jacob
3ed83fa681 * add Jacobi transformations 
* add Jacobi (Hestenes) SVD decomposition for square matrices
* add function for trivial Householder
 2009-08-09 16:58:13 +02:00
Gael Guennebaud
fe813911f2 make LU::solve() not to crash when rank=0 2009-08-09 00:06:53 +02:00
Benoit Jacob
5f8d58f36a fix bug in sorting of singular values 2009-08-09 00:05:38 +02:00
Thomas Capricelli
b10637be50 add basic .hgignore file for most common generated/temporary files 2009-08-08 21:42:14 +02:00
Gael Guennebaud
f5e1c896c7 replace custom rank one update in LU by an expression 2009-08-08 00:01:43 +02:00
Gael Guennebaud
d1dc088ef0 * implement a second level of micro blocking (faster for small sizes) 
* workaround GCC bad implementation of _mm_set1_p*
 2009-08-07 11:09:34 +02:00
Gael Guennebaud
543a785756 Fix compilation in sparse module 2009-08-06 17:28:49 +02:00
Gael Guennebaud
2707a6b87c fix determinant in PartialLU 2009-08-06 17:28:31 +02:00
Gael Guennebaud
1d1e4884da oops, one more bug fix in homogeneous 2009-08-06 16:56:10 +02:00
Gael Guennebaud
9822493aaf fixes in determinant and homogeneous 2009-08-06 16:54:55 +02:00
Gael Guennebaud
3ac01b1400 compilation fix in EigenSolver, 
bugfix in PartialLU
 2009-08-06 16:41:54 +02:00
Gael Guennebaud
e82e30862a typo 2009-08-06 15:04:42 +02:00
Gael Guennebaud
03febf00a0 fix VS compilation issue in MapBase::operator+= and -= 2009-08-06 15:03:37 +02:00
Gael Guennebaud
2e46e9f2b4 shame on me 2009-08-06 14:57:38 +02:00
Gael Guennebaud
d34c5ef509 fix my bad fix of Hauke's fix ;) 2009-08-06 14:54:25 +02:00
Gael Guennebaud
1d4fea48b5 fix a couple of compilations issues 2009-08-06 14:10:02 +02:00
Hauke Heibel
c2861dd41a fixed inversion for AffineCompact matrices 2009-08-06 12:25:18 +02:00
Gael Guennebaud
56d00779db more product refactoring 2009-08-06 12:20:02 +02:00
Hauke Heibel
6b2ab13ac5 fix vs.net compilation issue 2009-08-06 11:40:25 +02:00
Hauke Heibel
8163757cf0 fix vs.net compilation issue 2009-08-06 11:27:25 +02:00
Gael Guennebaud
fa55cf5ce7 fix compilation and segfault issues 2009-08-06 11:19:36 +02:00
Gael Guennebaud
b9b17ea5a5 add the missing Affine Transform * set of column vectors products... 2009-08-06 11:02:03 +02:00
Benoit Jacob
0744638b6f remove remnant of MultiplierBase 2009-08-06 10:35:13 +02:00
Gael Guennebaud
84a7659bef implement the missing outer product, 
and attempt to workaround a gcc internal error
 2009-08-05 17:39:11 +02:00
Gael Guennebaud
88147e0a91 big refactoring in Product.h: 
- all specialized products now inherits ProductBase
- the default product evaluated by Assign is still here,
  but it is currently enabled for small fixed sizes only
- => this significantly speed up compilation for large matrices
- I left the OuterProduct specialization empty as an exercise...
 2009-08-05 15:23:35 +02:00
Benoit Jacob
014c581a5b fix assertions, improve docs. 
we never assert on conditions that depend on the result of a computation!!
also the assertion that rank>0 amounts to matrix!=0 which we have to leave under the responsibility of the user.
 2009-08-05 10:15:28 +02:00
Benoit Jacob
b183a4f879 merge 2009-08-04 17:47:27 +02:00
Gael Guennebaud
ab6302376c merge 2009-08-04 17:12:00 +02:00