Mantis - ISSUES

0001026: alglib makes a wrong assumption for size of bool: is may not be 1 byte

The code assumes 1-byte bool, which is wrong for some platforms.

[ 26%] Building CXX object CMakeFiles/alglib.dir/src/ap.cpp.o
/usr/bin/g++-4.2 -Dalglib_EXPORTS -I/opt/local/var/macports/build/_opt_local_var_macports_sources_rsync.macports.org_macports_release_tarballs_ports_math_alglib/alglib/work/alglib-cpp/src -pipe -Os -DNDEBUG -I/opt/local/include -arch ppc -mmacosx-version-min=10.6 -fPIC -MD -MT CMakeFiles/alglib.dir/src/ap.cpp.o -MF CMakeFiles/alglib.dir/src/ap.cpp.o.d -o CMakeFiles/alglib.dir/src/ap.cpp.o -c /opt/local/var/macports/build/_opt_local_var_macports_sources_rsync.macports.org_macports_release_tarballs_ports_math_alglib/alglib/work/alglib-cpp/src/ap.cpp
/opt/local/var/macports/build/_opt_local_var_macports_sources_rsync.macports.org_macports_release_tarballs_ports_math_alglib/alglib/work/alglib-cpp/src/ap.cpp:294: error: size of array ‘_ae_bool_must_be_8_bits_wide’ is negative

0001025: alglib makes a wrong assumption for size of bool: is may not be 1 byte

0000947: SIMD support for Arm

Hello,

we wondered if you have plans to provide a NEON SIMD kernel implementation for arm processors? In particular we are interested in whether you plan to provide a SIMD implementation for Apple Silicon. Our software targets all desktop platforms. It would be useful to know if this is a feature we can expect in the future.

Kind regards,
Nathan

0000919: Default stopping conditions for MINBLEIC algorithm are too stringent

When passing in all 0s to minbleicsetcond, it is (according to documentation) supposed to lead to automatic stopping criterion selection. However, when this is done in the attached file, the result is a termination code of 7: "stopping conditions are too stringent, further improvement is impossible", making it difficult, for more complex cases, to determine if the issue is due to overly stringent automatic stopping criteria, or an issue with the problem setup that might also cause a significantly suboptimal point to be selected.

0000826: LM solver has to decrease step upon encountering infinity/nan

todo

0001037: test

test

0001036: ability to switch between 4-point and 2-point numerical diff

todo

0001035: spline1d batch evaluation/conversion for sorted points

todo: fullly orthogonal api, handle sorted inputs

0001030: FIXED: alignment-related bug in the HPC core (used by Java, Python, C#)

Under rare circumstances, the native HPC core experiences an alignment-related failure. It happens when the following conditions are met:
* a Linux system is used
* the application uses a non-standard malloc() implementation which produces insufficiently aligned memory blocks and which overrides ALGLIB choice of malloc() implementation.

Under these circumstances ALGLIB functions which erroneously use standard malloc() instead of ALGLIB's own aligned malloc may fail with SIGSEGV error.

0001029: IMPLEMENTED: incremental sparse matrix construction

New sparsecreatecrsempty() and sparseappendcompressedrow() functions which allow to subsequently add rows to an existing (and potentially empty) CRS-based matrix.

0001028: IMPLEMENTED: in-place multiplication of sparse matrix rows/cols

Implemented sparsemultiplycolsby() and sparsemultiplyrowsby()

0001027: IMPLEMENTED: new dense-to-sparse conversion functions

Implemented sparsecreatecrsfromdensev() and sparsecreatecrsfromdensevbuf(), which convert a dense M*N vector into a sparse MxN matrix stored in CRS format.

0001024: IMPLEMENTED: faster parallel sparse Cholesky/LDLT factorization, better parallelism in MINLP/MINQP

More fine-grained parallelism in sparse Cholesky/LDLT factorization results in better parallelism speed-up. Interior point MINLP/MINQP, which are heavily influenced by Cholesky performance, are also much faster now.

0001023: simulated annealing

todo, parallel

0000880: IMPLEMENTED: Lagrange multipliers for NLP

SQP solver now returns Lagrange multipliers.

0001019: multiobjective GDEMO

todo

0001018: multistart global optimization

todo

0001017: genetic algorithms

todo

0001012: hybrid DE+SQP solver

todo

0001007: minnlcsetlc2()

also various dense/sparse versions, "add" semantics

0001005: parallel callbacks for MINMO, MINNS

todo

0001001: parallel sparse gemv

todo

0001000: faster nipool, nrpool, nbpool

language support
thread unsafe retrieve functionality

0000977: remove _CRT_SECURE_NO_WARNINGS

no longer necessary

0000974: SQP improvements

* We need GammaMax in the new FSQP! rejected steps MUST contribute to GammaMax. We also need a special return code 99 which is intended for a situation when GammaMax is suspiciously large.

* We also need a positive completion code for a solver trapped in a nonzero constraint extremum.

* filter SQP has issues when starting from a local maximum of a constraints violation, like minnlc_d_equality with x0=[0,0]. An SL1QP is better, but adding L1 merit function is likely to overcomplicate the algo.