UBlas required?

[inducer]

Quick question: Is ViennaCL supposed to work without UBlas? (and without -DVIENNACL_WITH_UBLAS?)

Here's what I get:

x86_64-linux-gnu-gcc -pthread -fno-strict-aliasing -fwrapv -Wall -O3 -DNDEBUG -fPIC -Iexternal/boost_numpy/ -Iexternal/viennacl-dev/ -I/usr/lib/pymodules/python2.7/numpy/core/include -I/usr/include/python2.7 -c src/_viennacl/core.cpp -o build/temp.linux-x86_64-2.7/src/_viennacl/core.o
x86_64-linux-gnu-gcc -pthread -fno-strict-aliasing -fwrapv -Wall -O3 -DNDEBUG -fPIC -Iexternal/boost_numpy/ -Iexternal/viennacl-dev/ -I/usr/lib/pymodules/python2.7/numpy/core/include -I/usr/include/python2.7 -c src/_viennacl/vector_float.cpp -o build/temp.linux-x86_64-2.7/src/_viennacl/vector_float.o
In file included from src/_viennacl/vector_float.cpp:1:0:
src/_viennacl/vector.h: In instantiation of ‘viennacl::tools::shared_ptr<viennacl::vector<T> > vcl_vector_init_scalar(uint32_t, SCALARTYPE) [with SCALARTYPE = float; uint32_t = unsigned int]’:
src/_viennacl/vector_float.cpp:5:3:   required from here
src/_viennacl/vector.h:121:53: error: ‘boost::numeric::ublas::scalar_vector<float, std::allocator<float> > s_v’ has incomplete type
   ublas::scalar_vector<SCALARTYPE> s_v(length, value);

(Of course I'd like the answer to this to be a resounding 'yes'. :) )

[karlrupp]

Most parts of ViennaCL can also be used without uBLAS. From the error message it seems like this is not true for PyViennaCL. Maybe because Boost.Python is used anyway?

[tsmithe]

Hmm. It's fairly easy to remove this usage of ublas (and the parallel usage for matrix construction), but I also use the ublas sparse matrices as a host cache for constructing sparse matrix objects, and that's quite crucial. I really don't want to maintain my own sparse matrix implementation, just to avoid using ublas.

I'm not sure I understand the trepidation about ublas; as far as I can tell, you're trying to avoid boost altogether?

I'm also not so sure about using bpl-subset, but I'll discuss that in the other context.

Andreas Klöckner notifications@github.com writes:

Quick question: Is ViennaCL supposed to work without UBlas? (and without -DVIENNACL_WITH_UBLAS?)

Here's what I get:

x86_64-linux-gnu-gcc -pthread -fno-strict-aliasing -fwrapv -Wall -O3
-DNDEBUG -fPIC -Iexternal/boost_numpy/ -Iexternal/viennacl-dev/
-I/usr/lib/pymodules/python2.7/numpy/core/include
-I/usr/include/python2.7 -c src/_viennacl/core.cpp -o
build/temp.linux-x86_64-2.7/src/_viennacl/core.o
x86_64-linux-gnu-gcc -pthread -fno-strict-aliasing -fwrapv -Wall -O3
-DNDEBUG -fPIC -Iexternal/boost_numpy/ -Iexternal/viennacl-dev/
-I/usr/lib/pymodules/python2.7/numpy/core/include
-I/usr/include/python2.7 -c src/_viennacl/vector_float.cpp -o
build/temp.linux-x86_64-2.7/src/_viennacl/vector_float.o
In file included from src/_viennacl/vector_float.cpp:1:0:
src/_viennacl/vector.h: In instantiation of
‘viennacl::tools::shared_ptr<viennacl::vector<T> >
vcl_vector_init_scalar(uint32_t, SCALARTYPE) [with SCALARTYPE = float;
uint32_t = unsigned int]’:
src/_viennacl/vector_float.cpp:5:3:   required from here
src/_viennacl/vector.h:121:53: error: ‘boost::numeric::ublas::scalar_vector<float, std::allocator<float> > s_v’ has incomplete type
   ublas::scalar_vector<SCALARTYPE> s_v(length, value);

(Of course I'd like the answer to this to be a resounding 'yes'. :) )

---

Reply to this email directly or view it on GitHub: https://github.com/viennacl/pyviennacl-dev/issues/5