Conversation
|
Comment this line to skip the format check: Line 36 in 91028b1 |
|
Please rebase on the current master. I removed C++ 14. Therefore, it is less to test. |
996aa55 to
0fc20ee
Compare
|
Sorry for the delay, finally all checks have passed |
|
I see a conceptional problem in your implementation. If I understand it correctly, you pass a tuple of the begin pointers of each buffer ( std::tuple<bufType1, bufType2, bufType3> operator*(){
return std::make_tuple(std::get<0>(m_iteratorTuple[m_index]), std::get<1>(m_iteratorTuple[m_index]), std::get<2>(m_iteratorTuple[index]));
}
std::tuple<bufType1, bufType2, bufType3> operator[](int i){
return std::make_tuple(std::get<0>(m_iteratorTuple[i]), std::get<1>(m_iteratorTuple[i]), std::get<2>(m_iteratorTuple[i]));
}I think, the main problem is to make it generic for an arbitrary number of tuple members. I would also suggest, to test it first with auto transform = [](Data1 input) -> Data2 { return static_cast<Data2>(input);};
auto reduce = [](Data2 sum, Data2 input) -> Data2 { return sum + input;};Also the reduce functor is restricted to a single type for the moment, see here: #52 |
- rename was necessary to avoid confusions with fancy iterator - rename and flat namespace of the mem access strategy - move related files to new folder
- now, the memory access strategy returns the index instead the actual data element
- Change transform output from fancy iterator to a memory object of std::tuple<std::uint64_t, char, double> (like the return type of the functor) - Change transform result check by comparing element by element - Directly use the concrete types Update ZipIterator.hpp: - Remove = operator
- Fix operator[] - Change constexpr to ALPAKA_FN_HOST_ACC
50cccc6 to
53c7f10
Compare
Add:
Failed when building for accelerator testing