devtools/eigen
1
0
Fork 0
mirror of https://gitlab.com/libeigen/eigen.git synced 2026-04-10 11:34:33 +08:00
Code Issues Packages Projects Releases Wiki Activity
1,361 Commits 15 Branches 93 Tags
13e95f7f68f67ee0b66cfa37d9cdc9fdd71eeca8
Commit Graph

676 Commits

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
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
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
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
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
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
1d4fea48b5 fix a couple of compilations issues 2009-08-06 14:10:02 +02:00
Gael Guennebaud
56d00779db more product refactoring 2009-08-06 12:20:02 +02:00
Hauke Heibel
8163757cf0 fix vs.net compilation issue 2009-08-06 11:27:25 +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
Gael Guennebaud
7d607048a9 implement a ProductBase class and, as a proof of concept, update TriangularProduct 
and SelfAdjointMatrixProduct to take advantage of it => fewer LOC
 2009-08-04 16:54:17 +02:00
Benoit Jacob
523cdedf58 make the dot product linear in the second variable, not the first variable 2009-08-03 17:20:45 +02:00
Gael Guennebaud
912da9fade merge with special_matrix branch 2009-08-03 16:17:32 +02:00
Gael Guennebaud
a8f943127c merge 2009-08-03 16:11:30 +02:00
Benoit Jacob
d10c710b15 add new Householder module 2009-08-03 16:06:57 +02:00
Gael Guennebaud
3cf5bb31f6 * Bye bye MultiplierBase, extend a bit AnyMatrixBase to allow =, +=, and -= 
* This probably makes ReturnByValue needless
 2009-08-03 16:05:15 +02:00
Benoit Jacob
66ee2044ce small fixes 2009-08-03 16:05:07 +02:00
Benoit Jacob
3cde9c0e35 apply Gael's idea for auto transpose in mixed fixed/dynamic case 2009-08-03 16:04:15 +02:00
Gael Guennebaud
0103de8512 bugfix in trsm 2009-08-02 15:32:43 +02:00
Gael Guennebaud
48fc64458c add blocked LLT, and bugfix in trsm asserts 2009-08-01 23:42:51 +02:00
Gael Guennebaud
18429156a1 add selfadjointView from a trinagularView 2009-07-31 17:35:55 +02:00
Gael Guennebaud
a156f5a869 faster trsm kernel and fix a couple of issues 2009-07-31 13:18:19 +02:00
Gael Guennebaud
21f686846b s/std::atan2/ei_atan2 2009-07-31 10:08:23 +02:00
Manuel Yguel
ae5e26a363 add missing ei_atan2 without painfull warnings 2009-07-31 09:21:31 +02:00
Gael Guennebaud
ff20a2ba94 add explicit "on the right" triangular solving, 
=> no temporary when the rhs/unknows is row major
 2009-07-30 16:03:06 +02:00
Gael Guennebaud
62d9b9b7b5 fix typo 2009-07-29 09:26:20 +02:00
Gael Guennebaud
864171df5c fix a couple of issues related to recent products 2009-07-28 18:11:30 +02:00
Gael Guennebaud
1ba35248e9 synch with main branch 2009-07-28 17:37:22 +02:00
Gael Guennebaud
54804eb626 synch with main branch 2009-07-28 17:35:07 +02:00
Gael Guennebaud
264fe82c65 add a debug mechanism to compute the number of intermediate evaluations (only for dynamic size) 2009-07-28 17:13:13 +02:00
Gael Guennebaud
7ed7ec64b5 improve the expression analyzer to bypass Transpose expression 2009-07-28 14:02:12 +02:00
Gael Guennebaud
94cc30180e compilation fixes 2009-07-27 13:50:23 +02:00
Gael Guennebaud
0590c18555 various compilation and bug fixes in selfadjoint stuff 2009-07-27 13:17:39 +02:00
Gael Guennebaud
b5e4064289 cleaning and fix a perf issue 2009-07-27 12:13:53 +02:00