On Wednesday, October 19, 2016 at 7:03:49 AM UTC, Michele Zaffalon wrote: > > On Thursday, October 13, 2016 at 3:38:36 PM UTC+2, Steven G. Johnson wrote: >> >> write on a numeric array will output the raw bytes, i.e. Column-major >> data in the native byte order. >> >> >> Would it be a reasonable assumption that reshaping will not change the > ordering in future Julia implementations? >
[PyCall allows already changing to row-major, but there's a penalty then.] >From memory, I recall some talk of row-major could be choosable (in theory), can now only find: https://groups.google.com/forum/#!topic/julia-dev/Q-LFNapBdb0 "I like the way of Numpy of choose at creation time if order in memory has to be "fortran-style" (column-major) or "c-style" (row-major). If you now that operations are going to be intensive over columns, you can choice create a fortran-style matrix (and this avoid the time of create a c-style and do and transposition of the matrix)."