Hi. On Fri, Apr 04, 2014 at 03:46:48PM +0100, Barak A. Pearlmutter wrote: > Package: libarmadillo-dev > Version: 1:4.100.2+dfsg-2 > Severity: wishlist > > When compiling the mlpack machine learning library, which links to > Armadillo, the configuration scripts spits out this bon mot: > > -- Found Armadillo: /usr/lib/libarmadillo.so (found suitable version > "4.100.2", required is "3.6.0") > CMake Warning at CMakeLists.txt:81 (message): > This is a 64-bit system, but Armadillo was compiled without 64-bit index > support. Consider recompiling Armadillo with ARMA_64BIT_WORD to enable > 64-bit indices (large matrix support). MLPACK will still work without > ARMA_64BIT_WORD defined, but will not scale to matrices with more than 4 > billion elements. > > which sounds reasonable. > > At least, it sounds reasonable when you're doing machine learning with > big datasets. (I suppose there are other circumstances where 32-bit > indexes make more sense. But that argument would likely apply to all > pointers. Which would constitute a new 32-bit-user-pointers > amd64-instructions "architecture". Which is in fact under active > development. But it does seem hard to pretty hard to argue against > 64-bit index support on arch amd64.)
Sorry for the late response. I will try to see the best way to handle this. However, this could be handled at the mlpack level by just adding a #define appropriately before the armadillo include, for the time being. I will find out the best way to handle this and add it. Kumar -- Kumar Appaiah -- debian-science-maintainers mailing list [email protected] http://lists.alioth.debian.org/cgi-bin/mailman/listinfo/debian-science-maintainers
