Sure, and i will explain what the bug is.
Example:
typedef int vectorType; // just for convinience
cl::vector< int > devices(1,0); // sample vector, type doesnt matter
for (cl::vector<vectorType>::iterator device=devices.begin();device!=devices.end();++device)
std::cout << "prefix iterating devices..." << std::endl;
std::cout << "prefix iteration finished..." << std::endl;
for (cl::vector<vectorType>::iterator device=devices.begin();device!=devices.end();device++)
std::cout << "postfix iterating devices..." << std::endl;
std::cout << "postfix iteration finished..." << std::endl; // will not be reached
... will lock up in the second loop, i.e. looping as if the iterator would not be incremented at all. In fact, it does not get incremented at all.
Root of this evil is 'cl.hpp' Line 693ff (comments by me).
void operator++(int x) // looks like postfix ++
{
index_ += x; // OMG!
// according to standard postfix ops must return a new modified object
// additionally, value of x is undefined,
// so this one is -sorry- complete bs
}
Better:
iterator // this is a return value
operator ++(int x) // postfix increment
{
// assuming there is a copy constructor,
// wich in fact does not exist
iterator ret(*this);
index_++; // increment
return ret; // return copy,
// so any remaining code in the same expression still gets the old value
}
Conclusion:
Both postfix operators are bogus in two ways: a) x is undefined and likely zero b) even if x would be 1, the postfix operator would be a prefix operator.
Additionally the prefix ops are also not so standard
Some evil comments by me:
Those who coded cl.hpp obviously are C++ noobies. E.g. enforcing specific values for the various 'getInfo' methods could be better done with Enumerations instead of a template hell. Templates produce in most cases a code bloat. If there would be an speed advantage, there would be no problem (i really like templates). But there is absolutly no gain, the values get evaluated in OpenCL.dll/.so. A lot code is as mentioned roughly working but strange, i.e. not really how c++ code should look like. In fact, it looks like if some Java developers had an excursion to C++. And more...
Finally:
Have you any idea why i cant get any kernel compiled with >>GPU<< as target? (XP-SP3/VC90/Strem2.0-Beta4). It just returns Error -11.
Sample Binaries deployed by the official installer work fine, but sample project (HelloCL) build by me report error -11. Any own projects too, for GUI-based projects 3 console windows get openend before error -11 is reported, console projects just report error -11.
