The binop methods was determining whether to transpose or not based on is_standard_layout. These checks are replaced with fastest and slowest axis instead, closer matching the semantics wanted. This makes matrix slices cheaper to work with during addition and multiplication as we don't have to transpose a CSR into CSC unnecessarily
The
binop
methods was determining whether to transpose or not based onis_standard_layout
. These checks are replaced with fastest and slowest axis instead, closer matching the semantics wanted. This makes matrix slices cheaper to work with during addition and multiplication as we don't have to transpose aCSR
intoCSC
unnecessarily