63 Commits

Author SHA1 Message Date
Piotr Trojanek
12557fb2a2 more std:: namespace fixups for 2.0 branch 2010-05-01 09:25:36 +09:00
Piotr Trojanek
135013c608 EIGEN_ALIGN std:: fixup 2010-03-04 16:38:37 +01:00
Piotr Trojanek
1625a5e3f8 fixups for clean QCC compilation (add std:: in front of size_t, memcpy, etc.) 2010-03-05 11:21:11 +01:00
Piotr Trojanek
47a61bbd80 posix_memalign check for QNX 2010-03-04 17:12:30 +01:00
Benoit Jacob
f369dc873e Piotr's patch was missing many occurences of size_t. So,
using std::size_t;
This is the only way that we can ensure QCC support in the long term without having to think about it everytime.
2010-02-12 08:57:04 -05:00
Piotr Trojanek
7b44957c4b std:: namespace fixup for more restricive compilers such as QNX's QCC 2010-02-10 22:27:35 +01:00
Benoit Jacob
74207a31fa backport the fix to bug #79, and the unit test 2010-01-02 12:45:49 -05:00
Benoit Jacob
8084dbc86a copy the Memory.h file from the devel branch and remove some added trailing spaces.
This is now very harmless to do as the big change (EIGEN_ALIGN preprocessor stuff and the body of ei_aligned_malloc) was already introduced in 2.0.6.

Should address Björn's issue, and also improve FreeBSD platform detection.
2009-09-25 09:09:14 -04:00
Benoit Jacob
6af2c2c67a backported the following to 2.0:
* EIGEN_ALIGN and EIGEN_DONT_ALIGN and the corresponding logic in Macros.h
  (instead of using EIGEN_ARCH_WANTS_ALIGNMENT)
* The body of ei_aligned_malloc and ei_aligned_free

The reason for this backporting is that a user complained that with eigen 2.0 he got a warning at Memory.h:81 that the return value of posix_memalign was not used, and that function was declared with an attribute warn_unused_result.

Looking at this, it seemed that the body of this function was already overly complicated, and fixing this warning made it even worse, while the devel branch had a much simpler body and didn't suffer from that problem.

Then it was necessary to update ei_aligned_free too, and to backport EIGEN_ALIGN.

Inch' Allah....
2009-09-21 05:39:55 -04:00
Benoit Jacob
990615e884 backport 126284d08bcf9abb3a6665bf31305a188371e461
.
2009-07-31 13:30:12 +02:00
Gael Guennebaud
f5a167b3e7 apply patch from Hauke Heibel cleaning overloaded operator new/detete 2009-05-07 20:33:48 +00:00
Benoit Jacob
a29a390afa backport 958657: fix posix_memalign detection (Ross Smith) 2009-04-24 13:28:25 +00:00
Gael Guennebaud
c15842c374 backporting rev 951682 (compilation fix in aligned allocator) 2009-04-09 21:23:25 +00:00
Gael Guennebaud
d6bb34fa5a backporting various bug fixes related to MapBase/Map/Block and new
StdVector workaround because the previous was really too limited. I hope
it is not a too big change for a "stable" branch.
2009-03-24 08:20:43 +00:00
Benoit Jacob
d38504a4c8 backport 921254-921261 to the branch: disable alignment altogether on exotic platforms 2009-02-16 16:29:33 +00:00
Benoit Jacob
d2dcca52a3 backport 920106: BSD's don't have aligned malloc 2009-02-02 13:24:17 +00:00
Gael Guennebaud
9849931500 eventually it turns out that our current
EIGEN_WORK_AROUND_QT_BUG_CALLING_WRONG_OPERATOR_NEW_FIXED_IN_QT_4_5
is the right way to go for allowing placement new on a class having
overloaded operator new. Qt 4.5 won't add the :: prefix. (I still do not
understand how you can distinghish a placement new from an overloaded
operator new taking an allocator as argument...)
2009-01-23 16:31:03 +00:00
Benoit Jacob
5f43a42ee7 * remove set(), revert to old behavior where = resizes
* try to be clever in matrix ctors and operator=: be lazy when we can, always allow
  to copy rowvector into columnvector, check the template parameters,
  try to factor the code better
* add missing copy ctor in UnalignedType
* fix bug in the traits of DiagonalProduct
* renaming: EIGEN_TUNE_FOR_CPU_CACHE_SIZE
* update the dox a little
2009-01-21 17:10:23 +00:00
Benoit Jacob
385fd3d918 * clarify the situation with experimental parts
* remove all what was marked deprecated
2009-01-19 15:02:24 +00:00
Benoit Jacob
b179f8e1a4 add NetBSD to the list of OSes on which malloc is guaranteed to be 16
byte aligned, after discussion with Mark Davies.

CCMAIL: mark@ecs.vuw.ac.nz
2009-01-12 22:39:26 +00:00
Benoit Jacob
4d44ca226e * make std::vector specializations also for Transform and for Quaternion
* update test_stdvector
* Quaternion() does nothing (instead of bug)
* update test_geometry
* some renaming
2009-01-12 16:06:04 +00:00
Gael Guennebaud
f86818b5a6 bug fix in ei_stack_free 2009-01-12 14:20:21 +00:00
Gael Guennebaud
a4252584ed bugfix in ei_handmade_aligned_free for null pointers 2009-01-12 12:54:32 +00:00
Benoit Jacob
4361cb8913 EIGEN_NO_MALLOC must also block traditional unaligned malloc 2009-01-10 14:24:55 +00:00
Benoit Jacob
3efe6e4176 remove ei_new_allocator
remove corresponding part of test_dynalloc
2009-01-10 02:50:09 +00:00
Benoit Jacob
5bef59e371 (previous commit was bad) 2009-01-09 21:32:44 +00:00
Benoit Jacob
265ab86005 overloaded operator delete should call ei_conditinal_aligned_free, not
ei_aligned_free
2009-01-09 21:28:53 +00:00
Benoit Jacob
b3d580dec7 ei_aligned_delete was running through the various paths in the wrong
order
2009-01-09 20:57:06 +00:00
Benoit Jacob
fd831d5a12 * implement handmade aligned malloc, fast but always wastes 16 bytes of memory.
only used as fallback for now, needs benchmarking.
  also notice that some malloc() impls do waste memory to keep track of alignment
  and other stuff (check msdn's page on malloc).
* expand test_dynalloc to cover low level aligned alloc funcs. Remove the old
  #ifdef EIGEN_VECTORIZE...
* rewrite the logic choosing an aligned alloc, some new stuff:
  * malloc() already aligned on freebsd and windows x64 (plus apple already)
  * _mm_malloc() used only if EIGEN_VECTORIZE
  * posix_memalign: correct detection according to man page (not necessarily
    linux specific), don't attempt to declare it if the platform didn't declare it
    (there had to be a reason why it didn't declare it, right?)
2009-01-09 14:56:44 +00:00
Kenneth Frank Riddile
f52a9e5315 * Added aligned_allocator for using 16-byte aligned types with STL containers. There is still a compile-time problem with STL containers that have a standard-conformant resize() method, but this should resolve the original user issue which was storing aligned objects in a std::map. 2009-01-09 00:55:53 +00:00
Benoit Jacob
003d0ce03e add missing inline keywords (compilation error) spotted by timvdm 2009-01-08 20:20:11 +00:00
Benoit Jacob
51aa62a5d4 add EIGEN_WORK_AROUND_QT_BUG_CALLING_WRONG_OPERATOR_NEW_FIXED_IN_QT_4_5
an ugly hack that allows people to do QVector<Vector4f> with Qt <= 4.4
the Qt bug this is working around is fixed by Gael in Qt 4.5
2009-01-08 16:03:24 +00:00
Benoit Jacob
eb7dcbbfce EIGEN_MAKE_ALIGNED_OPERATOR_NEW didn't actually need to get the class
name as parameter
2009-01-08 15:37:13 +00:00
Benoit Jacob
1d52bd4cad the big memory changes. the most important changes are:
ei_aligned_malloc now really behaves like a malloc
 (untyped, doesn't call ctor)
ei_aligned_new is the typed variant calling ctor
EIGEN_MAKE_ALIGNED_OPERATOR_NEW now takes the class name as parameter
2009-01-08 15:20:21 +00:00
Gael Guennebaud
336f0a8486 fine tuning in dot() and sum(), and prepare for the sparse versions... 2009-01-07 16:58:17 +00:00
Armin Berres
8e888a45d0 call methods from the eigen namespace with prepended Eigen:: in define 2009-01-06 11:20:36 +00:00
Benoit Jacob
1c29d70312 * introduce macros to replace inheritance for operator new overloading
(former solution still available and tested)
  This plays much better with classes that already have base classes --
  don't force the user to mess with multiple inheritance, which gave
  much trouble with MSVC.
* Expand the unaligned assert dox page
* Minor fixes in the lazy evaluation dox page
2009-01-06 03:16:50 +00:00
Benoit Jacob
8551505436 problem solved, we really want public inheritance and it is only
automatic when the _child_ class is a struct.
2009-01-05 18:21:44 +00:00
Benoit Jacob
d316d4f393 fix compilation on apple: _mm_malloc was undefined. the fix is to just use malloc since on apple it already returns aligned ptrs 2009-01-05 13:15:27 +00:00
Benoit Jacob
15ca6659ac * the 4th template param of Matrix is now Options. One bit for storage
order, one bit for enabling/disabling auto-alignment. If you want to
disable, do:
Matrix<float,4,1,Matrix_DontAlign>
The Matrix_ prefix is the only way I can see to avoid
ambiguity/pollution. The old RowMajor, ColMajor constants are
deprecated, remain for now.
* this prompted several improvements in matrix_storage. ei_aligned_array
renamed to ei_matrix_array and moved there. The %16==0 tests are now
much more centralized in 1 place there.
* unalignedassert test: updated
* update FindEigen2.cmake from KDElibs
* determinant test: use VERIFY_IS_APPROX to fix false positives; add
testing of 1 big matrix
2009-01-04 15:26:32 +00:00
Benoit Jacob
d9e5fd393a * In LU solvers: no need anymore to use row-major matrices
* Matrix: always inherit WithAlignedOperatorNew, regardless of
vectorization or not
* rename ei_alloc_stack to ei_aligned_stack_alloc
* mixingtypes test: disable vectorization as SSE intrinsics don't allow
mixing types and we just get compile errors there.
2009-01-03 22:33:08 +00:00
Benoit Jacob
164f410bb5 * make WithAlignedOperatorNew always align even when vectorization is disabled
* make ei_aligned_malloc and ei_aligned_free honor custom operator new and delete
2008-12-30 14:11:35 +00:00
Benoit Jacob
361225068d fix bug discovered by Frank:
ei_alloc_stack must really return aligned pointers
2008-12-24 13:59:24 +00:00
Gael Guennebaud
5f6fbaa0e7 * fix a vectorization issue in Product
* use _mm_malloc/_mm_free on other platforms than linux of MSVC (eg., cygwin, OSX)
* replace a lot of inline keywords by EIGEN_STRONG_INLINE to compensate for
  poor MSVC inlining
2008-12-19 15:38:39 +00:00
Benoit Jacob
2110cca431 actually honor EIGEN_STACK_ALLOCATION.
Can set it to 0 to disable stack alloc which guarantees that bad alloc throws exceptions if they are enabled.
2008-12-16 15:44:48 +00:00
Benoit Jacob
38b83b4157 * throw bad_alloc if exceptions are enabled, after patch by Kenneth Riddile
* disable vectorization on MSVC 2005, as it doesn't have all the required intrinsics. require 2008.
2008-12-16 15:17:29 +00:00
Benoit Jacob
1ad751b991 only enable the "unaligned array" assert if vectorization is enabled.
if vectorization is disabled, WithAlignedOperatorNew is empty!
2008-12-15 20:49:03 +00:00
Benoit Jacob
55b603e457 * fix a bug I introduced in WithAlignedOperatorNew
* add an important comment
2008-12-15 20:14:59 +00:00
Benoit Jacob
763f0a2434 use ei_aligned_malloc and ei_aligned_free also in WithAlignedOperatorNew, so this too should now work with MSVC. 2008-12-15 17:55:11 +00:00
Benoit Jacob
9b1a3d6e19 small optimization (for MSVC) and simplification of ei_alligned_malloc 2008-12-15 17:33:19 +00:00