Hauke Heibel
3cc9e3f5bb
Fixes a compilation issue for MSVC.
2010-03-01 19:56:24 +01:00
Gael Guennebaud
65eba35f98
rm useless omp shared directive
2010-03-01 13:34:44 +01:00
Gael Guennebaud
1710c07f63
remove Qt's atomic dependency, I don't know what I was doing wrong...
2010-03-01 13:09:47 +01:00
Gael Guennebaud
31aa17e4ef
GEMM: move the first packing of A' before the packing of B'
2010-03-01 11:10:30 +01:00
Gael Guennebaud
aeff3ff391
make Aron's idea work using Qt's atomic implementation for the synchronisation
2010-03-01 10:57:32 +01:00
Benoit Jacob
f1f3c30ddc
remove the hack to make the static assertion on types actually show up.
...
indeed, now that we use the meta selector for transposing as needed, the static asserts work very well.
2010-02-28 11:10:13 -05:00
Benoit Jacob
07023b94d8
forgot defined(...)
2010-02-28 10:11:28 -05:00
Benoit Jacob
9334ed4444
on 64-bit systems, glibc's malloc returns 16-byte aligned ptrs, and we now take advantage of that.
2010-02-28 10:10:53 -05:00
Benoit Jacob
a480e7e60f
* fix ei_handmade_aligned_realloc (was calling realloc on wrong ptr)
...
* add missing std:: (at least for QNX compatibility)
* add big comments to "structure" the file
2010-02-28 09:10:41 -05:00
Hauke Heibel
ff8c2149c1
Added a generic reallocation implementation based on ei_aligned_malloc/_free.
...
Rewrote ei_handmade_aligned_realloc such that it is now using std::realloc.
Reorganized functions in Memory.h for better readability.
Add missing <cerrno> include to Core (it's now required in Memory.h).
2010-02-28 14:32:57 +01:00
Hauke Heibel
40bd69fbaa
Hide some internal stuff from the docs.
2010-02-28 12:56:37 +01:00
Benoit Jacob
1d9c18a8f3
comment out cerr's
2010-02-28 00:53:06 -05:00
Benoit Jacob
27f5250258
Only include <iosfwd> unless either EIGEN_DEBUG_ASSIGN is defined or we're in eigen2 support mode
2010-02-27 19:04:22 -05:00
Benoit Jacob
e84f7e07e9
add ei_posix_memalign_realloc
2010-02-27 18:57:07 -05:00
Benoit Jacob
22fabb8940
add missing inline keyword, thanks to Eamon.
2010-02-27 17:51:48 -05:00
Hauke Heibel
6c9eb36222
Added support for realloc based conservative resizing.
2010-02-27 17:25:07 +01:00
Hauke Heibel
78b2c7e16e
Fixed a typo.
2010-02-27 17:24:42 +01:00
Benoit Jacob
3f393490ad
dot: handle the rowvector.dot(colvector) case where one needs to transpose.
2010-02-27 11:19:14 -05:00
Benoit Jacob
d9f6380499
Remove the dot product's separate implementation and use cwiseProduct.sum instead.
...
Also take special care to get nicely working static assertions.
2010-02-27 10:03:27 -05:00
Benoit Jacob
2c9a91812e
merge
2010-02-26 21:47:54 -05:00
Benoit Jacob
814e40c72a
let redux use the new ByOuterInner accessors
2010-02-26 21:46:43 -05:00
Benoit Jacob
4927841cba
Document Map and Stride, add examples.
2010-02-26 21:29:04 -05:00
Benoit Jacob
b1f666d007
Fix Map-with-Stride and cover it by new unit tests.
2010-02-26 20:12:51 -05:00
Gael Guennebaud
6924bf2e99
implement Aron's idea of interleaving the packing with the first computations
2010-02-26 15:58:22 +01:00
nerbonne
c72a5074e6
Fixed perf problems for vector subtraction: inlining wasn't always happening when necessary.
2010-02-26 15:46:43 +01:00
Benoit Jacob
32115bff1e
* add VERIFY_IS_EQUAL, should compile faster and it's natural when no arithmetic is involved.
...
* rename 'submatrices' test to 'block'
* add block-inside-of-block tests
* remove old cruft
* split diagonal() tests into separate file
2010-02-26 09:03:13 -05:00
Gael Guennebaud
8d4a0e6753
fix compilation without openmp
2010-02-26 14:57:22 +01:00
Gael Guennebaud
3ac2b96a2f
implement a smarter parallelization strategy for gemm avoiding multiple
...
paking of the same data
2010-02-26 12:32:00 +01:00
Jitse Niesen
d86f5339b2
ComplexSchur: fix bug introduced in my previous commit.
...
The value of c is actually used a few lines later.
2010-02-26 09:47:17 +00:00
Benoit Jacob
f56ac04c34
DenseBase::IsRowMajor now takes the special case of vectors into account.
2010-02-25 21:24:42 -05:00
Benoit Jacob
b1c6c215a4
merge
2010-02-25 21:07:30 -05:00
Benoit Jacob
769641bc58
* Implement the ByOuterInner accessors
...
* use them (big simplification in Assign.h)
* axe (Inner|Outer)StrideAtCompileTime that were just introduced
* ei_int_if_dynamic now asserts that the size is the expected one: adapt to that in Block.h
* add rowStride() / colStride() in DenseBase
* implement innerStride() / outerStride() everywhere needed
2010-02-25 21:01:52 -05:00
Jitse Niesen
90e4a605ef
ComplexSchur: compute shift more stably, introduce exceptional shifts.
...
Both the new computation of the eigenvalues of a 2x2 block and the
exceptional shifts are taken from EISPACK routine COMQR.
2010-02-25 22:33:38 +00:00
Gael Guennebaud
53bae6b3f8
update matrix product selection rules for 1xSmallxLarge and the transposed case
2010-02-25 21:59:25 +01:00
Gael Guennebaud
959a1b5d63
detect and implement inplace permutations
2010-02-25 16:30:58 +01:00
Gael Guennebaud
d9ca0c0d36
optimize inverse permutations
2010-02-25 15:31:15 +01:00
Thomas Capricelli
0f3d69b65e
Provide "eigen" defines to decide which instruction set is used
...
(sse3, ssse3 and sse4), independantly from the compiler.
Only those defines should be used in other places, and the user can
rely on those to know which sets are used.
2010-02-24 21:43:30 +01:00
Gael Guennebaud
7c98c04412
add reconstructedMatrix() to LLT, and LUs
...
=> they show that some improvements have still to be done
for permutations, tr*tr, trapezoidal matrices
2010-02-24 19:16:10 +01:00
Gael Guennebaud
f7aa9873ca
* fix LDLT's default ctor use
...
* add a reconstructedMatrix() function to LDLT for debug purpose
2010-02-24 10:40:16 +01:00
Benoit Jacob
60325b8330
actually, this is not even meant to be a termination criterion. so the proper fix is this.
2010-02-23 16:10:26 -05:00
Benoit Jacob
3d066f4bc7
LDLT:
...
* fix bug thanks to Ben Goodrich: we were terminating at the wrong place, leaving some matrix coefficients with wrong values.
* don't use Higham's formula here: we're not trying to be rank-revealing.
2010-02-23 16:05:37 -05:00
Benoit Jacob
d92df336ad
Further LU test improvements. I'm not aware of any test failures anymore, not even with huge numbers of repetitions.
...
Finally the createRandomMatrixOfRank() function is renamed to createRandomPIMatrixOfRank, where PI stands for 'partial isometry', that is, a matrix whose singular values are 0 or 1.
2010-02-23 15:40:24 -05:00
Gael Guennebaud
a1e1103328
add a 2D parallelizer
2010-02-23 21:40:15 +01:00
Gael Guennebaud
022e2f5ef4
fix typo
2010-02-23 18:24:15 +01:00
Benoit Jacob
7dc75380c1
* FullPivLU: replace "remaining==0" termination condition (from Golub) by a fuzzy compare
...
(fixes lu test failures when testing solve())
* LU test: set appropriate threshold and limit the number of times that a specially tricky test
is run. (fixes lu test failures when testing rank()).
* Tests: rename createRandomMatrixOfRank to createRandomProjectionOfRank
2010-02-23 09:04:59 -05:00
Hauke Heibel
1fd8d7b96a
Attempt to fix PGI compilation issue.
2010-02-23 11:35:51 +01:00
Gael Guennebaud
d2b0eadf52
fully adapt the gebp kernel and optimize it for CPU with only 8 registers
...
(transplanted from 2ed88ebbf1995be90b8d0c25ff10248c8f56d023)
2010-02-22 16:35:05 +01:00
Gael Guennebaud
51a4b929a1
implement an even lower level version of the gebp kernel for MSVC (it seems to be faster with gcc as well)
...
(transplanted from 9a5643551fe068497f84a81cd8986febf1918382)
2010-02-22 15:18:29 +01:00
Gael Guennebaud
eb905500b6
significant speedup in the matrix-matrix products
2010-02-23 13:06:49 +01:00
Gael Guennebaud
fc4a85ecd5
fully adapt the gebp kernel and optimize it for CPU with only 8 registers
2010-02-22 16:35:05 +01:00
