MSVC 2022/c++20.
After employed c++20 language standard errors of compilation in intel oneTBB occurred.
To get rid of compilation errors proposed moving on actual (2021.5.0) intel oneTBB occ products and occt.
After migration errors of compilation of TKernel and TMath have been occurred...
...Error occurred if using c++20 standard with new oneTBB 2021.5.0.
The error was:
Error C2672 'tbb::v1::parallel_for_each': no matching overloaded function found TKernel
could be 'void tbb::detail::d2::parallel_for_each(Iterator,Iterator,const Body &)' TKernel
'tbb::detail::d2::parallel_for_each': the associated constraints are not satisfied TKernel
Note, that if we use c++14 or c++17, all ok, error does not occures.
To solve the problem, i have to modify `UniversalIterator` class:
`value_type` instead `UniversalIterator` converted to `IteratorInterface*`
`pointer` = `reference` = `value_type`
Method `DownCast` moved into `FunctorInterface` abstract class.
argument `UniversalIterator& item` of the unary fuctions converted to `IteratorInterface*`.
The proposed solution removes compilation error.
Affected projects: TKernel, TMath
Affected sources: src/OSD/OSD_Parallel.hxx, src/OSD/OSD_Parallel_Threads.cxx
Affected classes: class OSD_Parallel, OSD_Parallel::UniversalIterator, OSD_Parallel::FunctorInterface, OSD_Parallel::FunctorWrapperIter, OSD_Parallel:;FunctorWrapperInt.
Removed usage of std::auto_ptr.
opencascade::std workarounds for VS2008 have been dropped, when possible.
OCCT_NO_RVALUE_REFERENCE workaround for VS2008 has been removed.
OSD_Parallel::ToUseOcctThreads() - new flag allowing to use OCCT threads implementation even when compiled with TBB (for testing).
Added new command dparallel for managing default Thread Pool.
OSD_Parallel::For() now avoid creation of universal iterator in simplest case.
BOPTools_Parallel - eliminated redundant typedefs/explicit instantiations of templates.
Added functor using array of per-thread context instead of a map.
New class OSD_ThreadPool has been introduced to define a Thread Pool for multi-threading algorithm.
Thread Pool assigns a serial number for each thread allowing Multi-Threading algorithm to allocate thread-local storage variables as an array whose size is the same as the number of threads.
OSD_ThreadPool also redirects exceptions to a thread calling parallel execution and consistently initializes FPE exception handling.
New class Standard_Condition provides a platform-independent tool similar to Event in WinAPI.
A new auxiliary function Standard_Atomic_CompareAndSwap() has been introduced
for performing atomic compare and swap of integer number.
Standard_Atomic_Increment/Standard_Atomic_Decrement fallback implementation
using ASM code for x86 processors for GCC has been dropped;
instead, it is expected that GCC should be properly configured targeting modern x86 architectures.
OSD_Signal now declares fFltExceptions as thread_local variable accessible through OSD::ToCatchFloatingSignals() property.
Standard_THREADLOCAL macro (wrapping thread_local attribute) has been moved to public header Standard_Macro.hxx.
OSD_Parallel::ForEach() has been extended with new optional parameter theNbItems and uses OSD_ThreadPool::DefaultPool().
Implementation of methods OSD_Parallel::For() and ForEach() is moved to CXX files to avoid direct dependency of client code that uses OSD_Parallel on TBB headers, and necessity to link with TBB explicitly.
Runtime polymorphism (virtual methods) is used to hide implementation (TBB or threads-based).
Macro NO_CXX_EXCEPTION was removed from code.
Method Raise() was replaced by explicit throw statement.
Method Standard_Failure::Caught() was replaced by normal C++mechanism of exception transfer.
Method Standard_Failure::Caught() is deprecated now.
Eliminated empty constructors.
Updated samples.
Eliminate empty method ChangeValue from NCollection_Map class.
Removed not operable methods from NCollection classes.
Simple primitives to parallelize loops type "for" and "foreach" were implemented. The primitives encapsulates complete logic for creating and managing parallel context of loops. Moreover the primitives may be a wrapper for some primitives from 3rd-party library - TBB.
To use it is necessary to implement TBB like interface which is based on functors. For example:
Class Functor
{
public:
void operator() ([proccesing instance]) const
{
//...
}
};
In the body of the operator () should be implemented thread-safe logic of computations that can be performed in parallel context. If parallelized loop iterates on the collections with direct access by index (such as Vector, Array), it is more efficient to use the primitive ParallelFor (because it has no critical section).
All parts of OCC code which are using tbb were changed on new primitives.
0024826: Wrapping of parallelisation algorithms
Small fix.
