- 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...
=> faster a * b.transpose() product
=> this also fix a bug in a so far untested situation
* SYMM is now ready for use => still have to write the high level
stuff to convert natural expressions into a call to SYMM
handling conjuagted expression. still have to bench whether it
is faster (runtime and compile time) to directly call the
cache friendly functions, whence all the commented piece of code...
1 - make it easier to catch conjugate expressions
2 - make sure there is no unecessary copy (we had NestByValue<Derived> which seems to be very bad)
* update eigensolver wrt recent changes
* this allows to optimize xpr like C -= lazy_product, still have to catch "scalar_product_of_lazy_product"
* started to support conjugate in cache friendly products (very useful to evaluate A * B.adjoint() without
evaluating B.adjoint() into a temporary
* compilation fix
* move solveTriangular*() to TriangularView::solve*()
* move .llt() to SelfAdjointView
* add a high level wrapper to the efficient selfadjoint * vector product
* improve LLT so that we can specify which triangular part is meaningless
=> there are still many things to do (doc, cleaning, improve the matrix products, etc.)
introduce ei_is_diagonal to check for it
DiagonalCoeffs ---> Diagonal and allow Index to by Dynamic
-> add MatrixBase::diagonal(int) with unittest and doc
That means a lot of features which were available for sparse matrices
via the dense (and super slow) implemention are no longer available.
All features which make sense for sparse matrices (aka can be implemented efficiently) will be
implemented soon, but don't expect to see an API as rich as for the dense path.
Other changes:
* no block(), row(), col() anymore.
* instead use .innerVector() to get a col or row vector of a matrix.
* .segment(), start(), end() will be back soon, not sure for block()
* faster cwise product
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
* 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.
cachefriendlyproduct, that should be banned as well as depending on the
platform they can give a malloc, and they could happen even with (large
enough) fixed size matrices. Corresponding fix in Product.h:
cachefriendly is now only used for dynamic matrices -- fixedsize, no
matter how large, doesn't use the cachefriendly product. We don't need
to care (in my opinion) about performance for large fixed size, as large
fixed size is a bad idea in the first place and it is more important to
be able to guarantee clearly that fixed size never causes a malloc.
* fix issues in Product revealed by this test
* in Dot.h forbid mixing of different types (at least for now, might allow real.dot(complex) in the future).
* 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
