- Update to 0.16
* The commentary from 11/05/2018 was wrong. The constraints were not
reset to their natural order and remained randomized.
The other thing that was missing was full contraint reorder without
separation into independent and dependent ones. The algorithm doesn't
converge without it well.
* An incorrect optimization to Jacobian Copy building code from #1938
that resulted in corrupt data in multi-threaded execution mode was
fixed.
* An unintended change from commit #1898 has been reverted.
The QuickStep used to solve with randomized constraint order
each 8th iteration. The other iterations, the constraints
were reset to their natural order, as generated, with the dependent
constraints gathered in reverse order at end (the reverse order is
somehow important). With the commit #1898 the constraints were
randomly reordered each 8th iteration but then remained
in that randomized order and only were re-randomized on subsequent
multiples of 8.
* CMake support for project file generation has been added.
* dxHashSpace::collide() has been changed to fault host program
if scene gets too large and causes integer overflow.
* Memory and pointer size integer type use has been changed so that
internal typedefs are used instead of "_t" suffixed types.
* Introduction of cooperative algorithms API.
L*D*LT cooperative factorization and linear equation system
cooperative solving have been implemented.
* AtomicReadReorderBarrier, AtomicStore, AtomicStorePointer functions
have been added and some atomic function implementations have been
improved in OU.
- adjusted ode-iso-cpp.patch
- build with --enable-double-precision